Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/page_alloc.c | |
3 | * | |
4 | * Manages the free list, the system allocates free pages here. | |
5 | * Note that kmalloc() lives in slab.c | |
6 | * | |
7 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
8 | * Swap reorganised 29.12.95, Stephen Tweedie | |
9 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
10 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 | |
11 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
12 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 | |
13 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 | |
14 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/stddef.h> |
18 | #include <linux/mm.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/interrupt.h> | |
21 | #include <linux/pagemap.h> | |
22 | #include <linux/bootmem.h> | |
23 | #include <linux/compiler.h> | |
9f158333 | 24 | #include <linux/kernel.h> |
1da177e4 LT |
25 | #include <linux/module.h> |
26 | #include <linux/suspend.h> | |
27 | #include <linux/pagevec.h> | |
28 | #include <linux/blkdev.h> | |
29 | #include <linux/slab.h> | |
5a3135c2 | 30 | #include <linux/oom.h> |
1da177e4 LT |
31 | #include <linux/notifier.h> |
32 | #include <linux/topology.h> | |
33 | #include <linux/sysctl.h> | |
34 | #include <linux/cpu.h> | |
35 | #include <linux/cpuset.h> | |
bdc8cb98 | 36 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
37 | #include <linux/nodemask.h> |
38 | #include <linux/vmalloc.h> | |
4be38e35 | 39 | #include <linux/mempolicy.h> |
6811378e | 40 | #include <linux/stop_machine.h> |
c713216d MG |
41 | #include <linux/sort.h> |
42 | #include <linux/pfn.h> | |
3fcfab16 | 43 | #include <linux/backing-dev.h> |
933e312e | 44 | #include <linux/fault-inject.h> |
a5d76b54 | 45 | #include <linux/page-isolation.h> |
1da177e4 LT |
46 | |
47 | #include <asm/tlbflush.h> | |
ac924c60 | 48 | #include <asm/div64.h> |
1da177e4 LT |
49 | #include "internal.h" |
50 | ||
51 | /* | |
13808910 | 52 | * Array of node states. |
1da177e4 | 53 | */ |
13808910 CL |
54 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
55 | [N_POSSIBLE] = NODE_MASK_ALL, | |
56 | [N_ONLINE] = { { [0] = 1UL } }, | |
57 | #ifndef CONFIG_NUMA | |
58 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
59 | #ifdef CONFIG_HIGHMEM | |
60 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
61 | #endif | |
62 | [N_CPU] = { { [0] = 1UL } }, | |
63 | #endif /* NUMA */ | |
64 | }; | |
65 | EXPORT_SYMBOL(node_states); | |
66 | ||
6c231b7b | 67 | unsigned long totalram_pages __read_mostly; |
cb45b0e9 | 68 | unsigned long totalreserve_pages __read_mostly; |
1da177e4 | 69 | long nr_swap_pages; |
8ad4b1fb | 70 | int percpu_pagelist_fraction; |
1da177e4 | 71 | |
d9c23400 MG |
72 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
73 | int pageblock_order __read_mostly; | |
74 | #endif | |
75 | ||
d98c7a09 | 76 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 77 | |
1da177e4 LT |
78 | /* |
79 | * results with 256, 32 in the lowmem_reserve sysctl: | |
80 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
81 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
82 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
83 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
84 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
85 | * |
86 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
87 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 88 | */ |
2f1b6248 | 89 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
4b51d669 | 90 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 91 | 256, |
4b51d669 | 92 | #endif |
fb0e7942 | 93 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 94 | 256, |
fb0e7942 | 95 | #endif |
e53ef38d | 96 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 97 | 32, |
e53ef38d | 98 | #endif |
2a1e274a | 99 | 32, |
2f1b6248 | 100 | }; |
1da177e4 LT |
101 | |
102 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 | 103 | |
15ad7cdc | 104 | static char * const zone_names[MAX_NR_ZONES] = { |
4b51d669 | 105 | #ifdef CONFIG_ZONE_DMA |
2f1b6248 | 106 | "DMA", |
4b51d669 | 107 | #endif |
fb0e7942 | 108 | #ifdef CONFIG_ZONE_DMA32 |
2f1b6248 | 109 | "DMA32", |
fb0e7942 | 110 | #endif |
2f1b6248 | 111 | "Normal", |
e53ef38d | 112 | #ifdef CONFIG_HIGHMEM |
2a1e274a | 113 | "HighMem", |
e53ef38d | 114 | #endif |
2a1e274a | 115 | "Movable", |
2f1b6248 CL |
116 | }; |
117 | ||
1da177e4 LT |
118 | int min_free_kbytes = 1024; |
119 | ||
86356ab1 YG |
120 | unsigned long __meminitdata nr_kernel_pages; |
121 | unsigned long __meminitdata nr_all_pages; | |
a3142c8e | 122 | static unsigned long __meminitdata dma_reserve; |
1da177e4 | 123 | |
c713216d MG |
124 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
125 | /* | |
183ff22b | 126 | * MAX_ACTIVE_REGIONS determines the maximum number of distinct |
c713216d MG |
127 | * ranges of memory (RAM) that may be registered with add_active_range(). |
128 | * Ranges passed to add_active_range() will be merged if possible | |
129 | * so the number of times add_active_range() can be called is | |
130 | * related to the number of nodes and the number of holes | |
131 | */ | |
132 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
133 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
134 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
135 | #else | |
136 | #if MAX_NUMNODES >= 32 | |
137 | /* If there can be many nodes, allow up to 50 holes per node */ | |
138 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
139 | #else | |
140 | /* By default, allow up to 256 distinct regions */ | |
141 | #define MAX_ACTIVE_REGIONS 256 | |
142 | #endif | |
143 | #endif | |
144 | ||
98011f56 JB |
145 | static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS]; |
146 | static int __meminitdata nr_nodemap_entries; | |
147 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
148 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
fb01439c | 149 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
98011f56 JB |
150 | static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES]; |
151 | static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES]; | |
fb01439c | 152 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ |
2a1e274a | 153 | unsigned long __initdata required_kernelcore; |
484f51f8 | 154 | static unsigned long __initdata required_movablecore; |
e228929b | 155 | unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; |
2a1e274a MG |
156 | |
157 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
158 | int movable_zone; | |
159 | EXPORT_SYMBOL(movable_zone); | |
c713216d MG |
160 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
161 | ||
418508c1 MS |
162 | #if MAX_NUMNODES > 1 |
163 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
164 | EXPORT_SYMBOL(nr_node_ids); | |
165 | #endif | |
166 | ||
9ef9acb0 MG |
167 | int page_group_by_mobility_disabled __read_mostly; |
168 | ||
b2a0ac88 MG |
169 | static void set_pageblock_migratetype(struct page *page, int migratetype) |
170 | { | |
171 | set_pageblock_flags_group(page, (unsigned long)migratetype, | |
172 | PB_migrate, PB_migrate_end); | |
173 | } | |
174 | ||
13e7444b | 175 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 176 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 177 | { |
bdc8cb98 DH |
178 | int ret = 0; |
179 | unsigned seq; | |
180 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 181 | |
bdc8cb98 DH |
182 | do { |
183 | seq = zone_span_seqbegin(zone); | |
184 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
185 | ret = 1; | |
186 | else if (pfn < zone->zone_start_pfn) | |
187 | ret = 1; | |
188 | } while (zone_span_seqretry(zone, seq)); | |
189 | ||
190 | return ret; | |
c6a57e19 DH |
191 | } |
192 | ||
193 | static int page_is_consistent(struct zone *zone, struct page *page) | |
194 | { | |
14e07298 | 195 | if (!pfn_valid_within(page_to_pfn(page))) |
c6a57e19 | 196 | return 0; |
1da177e4 | 197 | if (zone != page_zone(page)) |
c6a57e19 DH |
198 | return 0; |
199 | ||
200 | return 1; | |
201 | } | |
202 | /* | |
203 | * Temporary debugging check for pages not lying within a given zone. | |
204 | */ | |
205 | static int bad_range(struct zone *zone, struct page *page) | |
206 | { | |
207 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 208 | return 1; |
c6a57e19 DH |
209 | if (!page_is_consistent(zone, page)) |
210 | return 1; | |
211 | ||
1da177e4 LT |
212 | return 0; |
213 | } | |
13e7444b NP |
214 | #else |
215 | static inline int bad_range(struct zone *zone, struct page *page) | |
216 | { | |
217 | return 0; | |
218 | } | |
219 | #endif | |
220 | ||
224abf92 | 221 | static void bad_page(struct page *page) |
1da177e4 | 222 | { |
224abf92 | 223 | printk(KERN_EMERG "Bad page state in process '%s'\n" |
7365f3d1 HD |
224 | KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" |
225 | KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
226 | KERN_EMERG "Backtrace:\n", | |
224abf92 NP |
227 | current->comm, page, (int)(2*sizeof(unsigned long)), |
228 | (unsigned long)page->flags, page->mapping, | |
229 | page_mapcount(page), page_count(page)); | |
1da177e4 | 230 | dump_stack(); |
334795ec HD |
231 | page->flags &= ~(1 << PG_lru | |
232 | 1 << PG_private | | |
1da177e4 | 233 | 1 << PG_locked | |
1da177e4 LT |
234 | 1 << PG_active | |
235 | 1 << PG_dirty | | |
334795ec HD |
236 | 1 << PG_reclaim | |
237 | 1 << PG_slab | | |
1da177e4 | 238 | 1 << PG_swapcache | |
676165a8 NP |
239 | 1 << PG_writeback | |
240 | 1 << PG_buddy ); | |
1da177e4 LT |
241 | set_page_count(page, 0); |
242 | reset_page_mapcount(page); | |
243 | page->mapping = NULL; | |
9f158333 | 244 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
245 | } |
246 | ||
1da177e4 LT |
247 | /* |
248 | * Higher-order pages are called "compound pages". They are structured thusly: | |
249 | * | |
250 | * The first PAGE_SIZE page is called the "head page". | |
251 | * | |
252 | * The remaining PAGE_SIZE pages are called "tail pages". | |
253 | * | |
254 | * All pages have PG_compound set. All pages have their ->private pointing at | |
255 | * the head page (even the head page has this). | |
256 | * | |
41d78ba5 HD |
257 | * The first tail page's ->lru.next holds the address of the compound page's |
258 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
259 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 260 | */ |
d98c7a09 HD |
261 | |
262 | static void free_compound_page(struct page *page) | |
263 | { | |
d85f3385 | 264 | __free_pages_ok(page, compound_order(page)); |
d98c7a09 HD |
265 | } |
266 | ||
1da177e4 LT |
267 | static void prep_compound_page(struct page *page, unsigned long order) |
268 | { | |
269 | int i; | |
270 | int nr_pages = 1 << order; | |
271 | ||
33f2ef89 | 272 | set_compound_page_dtor(page, free_compound_page); |
d85f3385 | 273 | set_compound_order(page, order); |
6d777953 | 274 | __SetPageHead(page); |
d85f3385 | 275 | for (i = 1; i < nr_pages; i++) { |
1da177e4 LT |
276 | struct page *p = page + i; |
277 | ||
d85f3385 | 278 | __SetPageTail(p); |
d85f3385 | 279 | p->first_page = page; |
1da177e4 LT |
280 | } |
281 | } | |
282 | ||
283 | static void destroy_compound_page(struct page *page, unsigned long order) | |
284 | { | |
285 | int i; | |
286 | int nr_pages = 1 << order; | |
287 | ||
d85f3385 | 288 | if (unlikely(compound_order(page) != order)) |
224abf92 | 289 | bad_page(page); |
1da177e4 | 290 | |
6d777953 | 291 | if (unlikely(!PageHead(page))) |
d85f3385 | 292 | bad_page(page); |
6d777953 | 293 | __ClearPageHead(page); |
d85f3385 | 294 | for (i = 1; i < nr_pages; i++) { |
1da177e4 LT |
295 | struct page *p = page + i; |
296 | ||
6d777953 | 297 | if (unlikely(!PageTail(p) | |
d85f3385 | 298 | (p->first_page != page))) |
224abf92 | 299 | bad_page(page); |
d85f3385 | 300 | __ClearPageTail(p); |
1da177e4 LT |
301 | } |
302 | } | |
1da177e4 | 303 | |
17cf4406 NP |
304 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
305 | { | |
306 | int i; | |
307 | ||
6626c5d5 AM |
308 | /* |
309 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
310 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
311 | */ | |
725d704e | 312 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
17cf4406 NP |
313 | for (i = 0; i < (1 << order); i++) |
314 | clear_highpage(page + i); | |
315 | } | |
316 | ||
6aa3001b AM |
317 | static inline void set_page_order(struct page *page, int order) |
318 | { | |
4c21e2f2 | 319 | set_page_private(page, order); |
676165a8 | 320 | __SetPageBuddy(page); |
1da177e4 LT |
321 | } |
322 | ||
323 | static inline void rmv_page_order(struct page *page) | |
324 | { | |
676165a8 | 325 | __ClearPageBuddy(page); |
4c21e2f2 | 326 | set_page_private(page, 0); |
1da177e4 LT |
327 | } |
328 | ||
329 | /* | |
330 | * Locate the struct page for both the matching buddy in our | |
331 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
332 | * | |
333 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
334 | * the following equation: | |
335 | * B2 = B1 ^ (1 << O) | |
336 | * For example, if the starting buddy (buddy2) is #8 its order | |
337 | * 1 buddy is #10: | |
338 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
339 | * | |
340 | * 2) Any buddy B will have an order O+1 parent P which | |
341 | * satisfies the following equation: | |
342 | * P = B & ~(1 << O) | |
343 | * | |
d6e05edc | 344 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
1da177e4 LT |
345 | */ |
346 | static inline struct page * | |
347 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
348 | { | |
349 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
350 | ||
351 | return page + (buddy_idx - page_idx); | |
352 | } | |
353 | ||
354 | static inline unsigned long | |
355 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
356 | { | |
357 | return (page_idx & ~(1 << order)); | |
358 | } | |
359 | ||
360 | /* | |
361 | * This function checks whether a page is free && is the buddy | |
362 | * we can do coalesce a page and its buddy if | |
13e7444b | 363 | * (a) the buddy is not in a hole && |
676165a8 | 364 | * (b) the buddy is in the buddy system && |
cb2b95e1 AW |
365 | * (c) a page and its buddy have the same order && |
366 | * (d) a page and its buddy are in the same zone. | |
676165a8 NP |
367 | * |
368 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
369 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
1da177e4 | 370 | * |
676165a8 | 371 | * For recording page's order, we use page_private(page). |
1da177e4 | 372 | */ |
cb2b95e1 AW |
373 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
374 | int order) | |
1da177e4 | 375 | { |
14e07298 | 376 | if (!pfn_valid_within(page_to_pfn(buddy))) |
13e7444b | 377 | return 0; |
13e7444b | 378 | |
cb2b95e1 AW |
379 | if (page_zone_id(page) != page_zone_id(buddy)) |
380 | return 0; | |
381 | ||
382 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
383 | BUG_ON(page_count(buddy) != 0); | |
6aa3001b | 384 | return 1; |
676165a8 | 385 | } |
6aa3001b | 386 | return 0; |
1da177e4 LT |
387 | } |
388 | ||
389 | /* | |
390 | * Freeing function for a buddy system allocator. | |
391 | * | |
392 | * The concept of a buddy system is to maintain direct-mapped table | |
393 | * (containing bit values) for memory blocks of various "orders". | |
394 | * The bottom level table contains the map for the smallest allocatable | |
395 | * units of memory (here, pages), and each level above it describes | |
396 | * pairs of units from the levels below, hence, "buddies". | |
397 | * At a high level, all that happens here is marking the table entry | |
398 | * at the bottom level available, and propagating the changes upward | |
399 | * as necessary, plus some accounting needed to play nicely with other | |
400 | * parts of the VM system. | |
401 | * At each level, we keep a list of pages, which are heads of continuous | |
676165a8 | 402 | * free pages of length of (1 << order) and marked with PG_buddy. Page's |
4c21e2f2 | 403 | * order is recorded in page_private(page) field. |
1da177e4 LT |
404 | * So when we are allocating or freeing one, we can derive the state of the |
405 | * other. That is, if we allocate a small block, and both were | |
406 | * free, the remainder of the region must be split into blocks. | |
407 | * If a block is freed, and its buddy is also free, then this | |
408 | * triggers coalescing into a block of larger size. | |
409 | * | |
410 | * -- wli | |
411 | */ | |
412 | ||
48db57f8 | 413 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
414 | struct zone *zone, unsigned int order) |
415 | { | |
416 | unsigned long page_idx; | |
417 | int order_size = 1 << order; | |
b2a0ac88 | 418 | int migratetype = get_pageblock_migratetype(page); |
1da177e4 | 419 | |
224abf92 | 420 | if (unlikely(PageCompound(page))) |
1da177e4 LT |
421 | destroy_compound_page(page, order); |
422 | ||
423 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
424 | ||
725d704e NP |
425 | VM_BUG_ON(page_idx & (order_size - 1)); |
426 | VM_BUG_ON(bad_range(zone, page)); | |
1da177e4 | 427 | |
d23ad423 | 428 | __mod_zone_page_state(zone, NR_FREE_PAGES, order_size); |
1da177e4 LT |
429 | while (order < MAX_ORDER-1) { |
430 | unsigned long combined_idx; | |
1da177e4 LT |
431 | struct page *buddy; |
432 | ||
1da177e4 | 433 | buddy = __page_find_buddy(page, page_idx, order); |
cb2b95e1 | 434 | if (!page_is_buddy(page, buddy, order)) |
1da177e4 | 435 | break; /* Move the buddy up one level. */ |
13e7444b | 436 | |
1da177e4 | 437 | list_del(&buddy->lru); |
b2a0ac88 | 438 | zone->free_area[order].nr_free--; |
1da177e4 | 439 | rmv_page_order(buddy); |
13e7444b | 440 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
441 | page = page + (combined_idx - page_idx); |
442 | page_idx = combined_idx; | |
443 | order++; | |
444 | } | |
445 | set_page_order(page, order); | |
b2a0ac88 MG |
446 | list_add(&page->lru, |
447 | &zone->free_area[order].free_list[migratetype]); | |
1da177e4 LT |
448 | zone->free_area[order].nr_free++; |
449 | } | |
450 | ||
224abf92 | 451 | static inline int free_pages_check(struct page *page) |
1da177e4 | 452 | { |
92be2e33 NP |
453 | if (unlikely(page_mapcount(page) | |
454 | (page->mapping != NULL) | | |
455 | (page_count(page) != 0) | | |
1da177e4 LT |
456 | (page->flags & ( |
457 | 1 << PG_lru | | |
458 | 1 << PG_private | | |
459 | 1 << PG_locked | | |
460 | 1 << PG_active | | |
1da177e4 LT |
461 | 1 << PG_slab | |
462 | 1 << PG_swapcache | | |
b5810039 | 463 | 1 << PG_writeback | |
676165a8 NP |
464 | 1 << PG_reserved | |
465 | 1 << PG_buddy )))) | |
224abf92 | 466 | bad_page(page); |
1da177e4 | 467 | if (PageDirty(page)) |
242e5468 | 468 | __ClearPageDirty(page); |
689bcebf HD |
469 | /* |
470 | * For now, we report if PG_reserved was found set, but do not | |
471 | * clear it, and do not free the page. But we shall soon need | |
472 | * to do more, for when the ZERO_PAGE count wraps negative. | |
473 | */ | |
474 | return PageReserved(page); | |
1da177e4 LT |
475 | } |
476 | ||
477 | /* | |
478 | * Frees a list of pages. | |
479 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 480 | * count is the number of pages to free. |
1da177e4 LT |
481 | * |
482 | * If the zone was previously in an "all pages pinned" state then look to | |
483 | * see if this freeing clears that state. | |
484 | * | |
485 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
486 | * pinned" detection logic. | |
487 | */ | |
48db57f8 NP |
488 | static void free_pages_bulk(struct zone *zone, int count, |
489 | struct list_head *list, int order) | |
1da177e4 | 490 | { |
c54ad30c | 491 | spin_lock(&zone->lock); |
e815af95 | 492 | zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE); |
1da177e4 | 493 | zone->pages_scanned = 0; |
48db57f8 NP |
494 | while (count--) { |
495 | struct page *page; | |
496 | ||
725d704e | 497 | VM_BUG_ON(list_empty(list)); |
1da177e4 | 498 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 499 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 500 | list_del(&page->lru); |
48db57f8 | 501 | __free_one_page(page, zone, order); |
1da177e4 | 502 | } |
c54ad30c | 503 | spin_unlock(&zone->lock); |
1da177e4 LT |
504 | } |
505 | ||
48db57f8 | 506 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 | 507 | { |
006d22d9 | 508 | spin_lock(&zone->lock); |
e815af95 | 509 | zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE); |
006d22d9 | 510 | zone->pages_scanned = 0; |
0798e519 | 511 | __free_one_page(page, zone, order); |
006d22d9 | 512 | spin_unlock(&zone->lock); |
48db57f8 NP |
513 | } |
514 | ||
515 | static void __free_pages_ok(struct page *page, unsigned int order) | |
516 | { | |
517 | unsigned long flags; | |
1da177e4 | 518 | int i; |
689bcebf | 519 | int reserved = 0; |
1da177e4 | 520 | |
1da177e4 | 521 | for (i = 0 ; i < (1 << order) ; ++i) |
224abf92 | 522 | reserved += free_pages_check(page + i); |
689bcebf HD |
523 | if (reserved) |
524 | return; | |
525 | ||
9858db50 NP |
526 | if (!PageHighMem(page)) |
527 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); | |
dafb1367 | 528 | arch_free_page(page, order); |
48db57f8 | 529 | kernel_map_pages(page, 1 << order, 0); |
dafb1367 | 530 | |
c54ad30c | 531 | local_irq_save(flags); |
f8891e5e | 532 | __count_vm_events(PGFREE, 1 << order); |
48db57f8 | 533 | free_one_page(page_zone(page), page, order); |
c54ad30c | 534 | local_irq_restore(flags); |
1da177e4 LT |
535 | } |
536 | ||
a226f6c8 DH |
537 | /* |
538 | * permit the bootmem allocator to evade page validation on high-order frees | |
539 | */ | |
920c7a5d | 540 | void __init __free_pages_bootmem(struct page *page, unsigned int order) |
a226f6c8 DH |
541 | { |
542 | if (order == 0) { | |
543 | __ClearPageReserved(page); | |
544 | set_page_count(page, 0); | |
7835e98b | 545 | set_page_refcounted(page); |
545b1ea9 | 546 | __free_page(page); |
a226f6c8 | 547 | } else { |
a226f6c8 DH |
548 | int loop; |
549 | ||
545b1ea9 | 550 | prefetchw(page); |
a226f6c8 DH |
551 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
552 | struct page *p = &page[loop]; | |
553 | ||
545b1ea9 NP |
554 | if (loop + 1 < BITS_PER_LONG) |
555 | prefetchw(p + 1); | |
a226f6c8 DH |
556 | __ClearPageReserved(p); |
557 | set_page_count(p, 0); | |
558 | } | |
559 | ||
7835e98b | 560 | set_page_refcounted(page); |
545b1ea9 | 561 | __free_pages(page, order); |
a226f6c8 DH |
562 | } |
563 | } | |
564 | ||
1da177e4 LT |
565 | |
566 | /* | |
567 | * The order of subdivision here is critical for the IO subsystem. | |
568 | * Please do not alter this order without good reasons and regression | |
569 | * testing. Specifically, as large blocks of memory are subdivided, | |
570 | * the order in which smaller blocks are delivered depends on the order | |
571 | * they're subdivided in this function. This is the primary factor | |
572 | * influencing the order in which pages are delivered to the IO | |
573 | * subsystem according to empirical testing, and this is also justified | |
574 | * by considering the behavior of a buddy system containing a single | |
575 | * large block of memory acted on by a series of small allocations. | |
576 | * This behavior is a critical factor in sglist merging's success. | |
577 | * | |
578 | * -- wli | |
579 | */ | |
085cc7d5 | 580 | static inline void expand(struct zone *zone, struct page *page, |
b2a0ac88 MG |
581 | int low, int high, struct free_area *area, |
582 | int migratetype) | |
1da177e4 LT |
583 | { |
584 | unsigned long size = 1 << high; | |
585 | ||
586 | while (high > low) { | |
587 | area--; | |
588 | high--; | |
589 | size >>= 1; | |
725d704e | 590 | VM_BUG_ON(bad_range(zone, &page[size])); |
b2a0ac88 | 591 | list_add(&page[size].lru, &area->free_list[migratetype]); |
1da177e4 LT |
592 | area->nr_free++; |
593 | set_page_order(&page[size], high); | |
594 | } | |
1da177e4 LT |
595 | } |
596 | ||
1da177e4 LT |
597 | /* |
598 | * This page is about to be returned from the page allocator | |
599 | */ | |
17cf4406 | 600 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 601 | { |
92be2e33 NP |
602 | if (unlikely(page_mapcount(page) | |
603 | (page->mapping != NULL) | | |
604 | (page_count(page) != 0) | | |
334795ec HD |
605 | (page->flags & ( |
606 | 1 << PG_lru | | |
1da177e4 LT |
607 | 1 << PG_private | |
608 | 1 << PG_locked | | |
1da177e4 LT |
609 | 1 << PG_active | |
610 | 1 << PG_dirty | | |
334795ec | 611 | 1 << PG_slab | |
1da177e4 | 612 | 1 << PG_swapcache | |
b5810039 | 613 | 1 << PG_writeback | |
676165a8 NP |
614 | 1 << PG_reserved | |
615 | 1 << PG_buddy )))) | |
224abf92 | 616 | bad_page(page); |
1da177e4 | 617 | |
689bcebf HD |
618 | /* |
619 | * For now, we report if PG_reserved was found set, but do not | |
620 | * clear it, and do not allocate the page: as a safety net. | |
621 | */ | |
622 | if (PageReserved(page)) | |
623 | return 1; | |
624 | ||
d77c2d7c | 625 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead | |
1da177e4 | 626 | 1 << PG_referenced | 1 << PG_arch_1 | |
5409bae0 | 627 | 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk); |
4c21e2f2 | 628 | set_page_private(page, 0); |
7835e98b | 629 | set_page_refcounted(page); |
cc102509 NP |
630 | |
631 | arch_alloc_page(page, order); | |
1da177e4 | 632 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
633 | |
634 | if (gfp_flags & __GFP_ZERO) | |
635 | prep_zero_page(page, order, gfp_flags); | |
636 | ||
637 | if (order && (gfp_flags & __GFP_COMP)) | |
638 | prep_compound_page(page, order); | |
639 | ||
689bcebf | 640 | return 0; |
1da177e4 LT |
641 | } |
642 | ||
56fd56b8 MG |
643 | /* |
644 | * Go through the free lists for the given migratetype and remove | |
645 | * the smallest available page from the freelists | |
646 | */ | |
647 | static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
648 | int migratetype) | |
649 | { | |
650 | unsigned int current_order; | |
651 | struct free_area * area; | |
652 | struct page *page; | |
653 | ||
654 | /* Find a page of the appropriate size in the preferred list */ | |
655 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
656 | area = &(zone->free_area[current_order]); | |
657 | if (list_empty(&area->free_list[migratetype])) | |
658 | continue; | |
659 | ||
660 | page = list_entry(area->free_list[migratetype].next, | |
661 | struct page, lru); | |
662 | list_del(&page->lru); | |
663 | rmv_page_order(page); | |
664 | area->nr_free--; | |
665 | __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order)); | |
666 | expand(zone, page, order, current_order, area, migratetype); | |
667 | return page; | |
668 | } | |
669 | ||
670 | return NULL; | |
671 | } | |
672 | ||
673 | ||
b2a0ac88 MG |
674 | /* |
675 | * This array describes the order lists are fallen back to when | |
676 | * the free lists for the desirable migrate type are depleted | |
677 | */ | |
678 | static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = { | |
64c5e135 MG |
679 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, |
680 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
681 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
682 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */ | |
b2a0ac88 MG |
683 | }; |
684 | ||
c361be55 MG |
685 | /* |
686 | * Move the free pages in a range to the free lists of the requested type. | |
d9c23400 | 687 | * Note that start_page and end_pages are not aligned on a pageblock |
c361be55 MG |
688 | * boundary. If alignment is required, use move_freepages_block() |
689 | */ | |
690 | int move_freepages(struct zone *zone, | |
691 | struct page *start_page, struct page *end_page, | |
692 | int migratetype) | |
693 | { | |
694 | struct page *page; | |
695 | unsigned long order; | |
d100313f | 696 | int pages_moved = 0; |
c361be55 MG |
697 | |
698 | #ifndef CONFIG_HOLES_IN_ZONE | |
699 | /* | |
700 | * page_zone is not safe to call in this context when | |
701 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
702 | * anyway as we check zone boundaries in move_freepages_block(). | |
703 | * Remove at a later date when no bug reports exist related to | |
ac0e5b7a | 704 | * grouping pages by mobility |
c361be55 MG |
705 | */ |
706 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
707 | #endif | |
708 | ||
709 | for (page = start_page; page <= end_page;) { | |
710 | if (!pfn_valid_within(page_to_pfn(page))) { | |
711 | page++; | |
712 | continue; | |
713 | } | |
714 | ||
715 | if (!PageBuddy(page)) { | |
716 | page++; | |
717 | continue; | |
718 | } | |
719 | ||
720 | order = page_order(page); | |
721 | list_del(&page->lru); | |
722 | list_add(&page->lru, | |
723 | &zone->free_area[order].free_list[migratetype]); | |
724 | page += 1 << order; | |
d100313f | 725 | pages_moved += 1 << order; |
c361be55 MG |
726 | } |
727 | ||
d100313f | 728 | return pages_moved; |
c361be55 MG |
729 | } |
730 | ||
731 | int move_freepages_block(struct zone *zone, struct page *page, int migratetype) | |
732 | { | |
733 | unsigned long start_pfn, end_pfn; | |
734 | struct page *start_page, *end_page; | |
735 | ||
736 | start_pfn = page_to_pfn(page); | |
d9c23400 | 737 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
c361be55 | 738 | start_page = pfn_to_page(start_pfn); |
d9c23400 MG |
739 | end_page = start_page + pageblock_nr_pages - 1; |
740 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
c361be55 MG |
741 | |
742 | /* Do not cross zone boundaries */ | |
743 | if (start_pfn < zone->zone_start_pfn) | |
744 | start_page = page; | |
745 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
746 | return 0; | |
747 | ||
748 | return move_freepages(zone, start_page, end_page, migratetype); | |
749 | } | |
750 | ||
b2a0ac88 MG |
751 | /* Remove an element from the buddy allocator from the fallback list */ |
752 | static struct page *__rmqueue_fallback(struct zone *zone, int order, | |
753 | int start_migratetype) | |
754 | { | |
755 | struct free_area * area; | |
756 | int current_order; | |
757 | struct page *page; | |
758 | int migratetype, i; | |
759 | ||
760 | /* Find the largest possible block of pages in the other list */ | |
761 | for (current_order = MAX_ORDER-1; current_order >= order; | |
762 | --current_order) { | |
763 | for (i = 0; i < MIGRATE_TYPES - 1; i++) { | |
764 | migratetype = fallbacks[start_migratetype][i]; | |
765 | ||
56fd56b8 MG |
766 | /* MIGRATE_RESERVE handled later if necessary */ |
767 | if (migratetype == MIGRATE_RESERVE) | |
768 | continue; | |
e010487d | 769 | |
b2a0ac88 MG |
770 | area = &(zone->free_area[current_order]); |
771 | if (list_empty(&area->free_list[migratetype])) | |
772 | continue; | |
773 | ||
774 | page = list_entry(area->free_list[migratetype].next, | |
775 | struct page, lru); | |
776 | area->nr_free--; | |
777 | ||
778 | /* | |
c361be55 | 779 | * If breaking a large block of pages, move all free |
46dafbca MG |
780 | * pages to the preferred allocation list. If falling |
781 | * back for a reclaimable kernel allocation, be more | |
782 | * agressive about taking ownership of free pages | |
b2a0ac88 | 783 | */ |
d9c23400 | 784 | if (unlikely(current_order >= (pageblock_order >> 1)) || |
46dafbca MG |
785 | start_migratetype == MIGRATE_RECLAIMABLE) { |
786 | unsigned long pages; | |
787 | pages = move_freepages_block(zone, page, | |
788 | start_migratetype); | |
789 | ||
790 | /* Claim the whole block if over half of it is free */ | |
d9c23400 | 791 | if (pages >= (1 << (pageblock_order-1))) |
46dafbca MG |
792 | set_pageblock_migratetype(page, |
793 | start_migratetype); | |
794 | ||
b2a0ac88 | 795 | migratetype = start_migratetype; |
c361be55 | 796 | } |
b2a0ac88 MG |
797 | |
798 | /* Remove the page from the freelists */ | |
799 | list_del(&page->lru); | |
800 | rmv_page_order(page); | |
801 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
802 | -(1UL << order)); | |
803 | ||
d9c23400 | 804 | if (current_order == pageblock_order) |
b2a0ac88 MG |
805 | set_pageblock_migratetype(page, |
806 | start_migratetype); | |
807 | ||
808 | expand(zone, page, order, current_order, area, migratetype); | |
809 | return page; | |
810 | } | |
811 | } | |
812 | ||
56fd56b8 MG |
813 | /* Use MIGRATE_RESERVE rather than fail an allocation */ |
814 | return __rmqueue_smallest(zone, order, MIGRATE_RESERVE); | |
b2a0ac88 MG |
815 | } |
816 | ||
56fd56b8 | 817 | /* |
1da177e4 LT |
818 | * Do the hard work of removing an element from the buddy allocator. |
819 | * Call me with the zone->lock already held. | |
820 | */ | |
b2a0ac88 MG |
821 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
822 | int migratetype) | |
1da177e4 | 823 | { |
1da177e4 LT |
824 | struct page *page; |
825 | ||
56fd56b8 | 826 | page = __rmqueue_smallest(zone, order, migratetype); |
b2a0ac88 | 827 | |
56fd56b8 MG |
828 | if (unlikely(!page)) |
829 | page = __rmqueue_fallback(zone, order, migratetype); | |
b2a0ac88 MG |
830 | |
831 | return page; | |
1da177e4 LT |
832 | } |
833 | ||
834 | /* | |
835 | * Obtain a specified number of elements from the buddy allocator, all under | |
836 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
837 | * Returns the number of new pages which were placed at *list. | |
838 | */ | |
839 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
b2a0ac88 MG |
840 | unsigned long count, struct list_head *list, |
841 | int migratetype) | |
1da177e4 | 842 | { |
1da177e4 | 843 | int i; |
1da177e4 | 844 | |
c54ad30c | 845 | spin_lock(&zone->lock); |
1da177e4 | 846 | for (i = 0; i < count; ++i) { |
b2a0ac88 | 847 | struct page *page = __rmqueue(zone, order, migratetype); |
085cc7d5 | 848 | if (unlikely(page == NULL)) |
1da177e4 | 849 | break; |
81eabcbe MG |
850 | |
851 | /* | |
852 | * Split buddy pages returned by expand() are received here | |
853 | * in physical page order. The page is added to the callers and | |
854 | * list and the list head then moves forward. From the callers | |
855 | * perspective, the linked list is ordered by page number in | |
856 | * some conditions. This is useful for IO devices that can | |
857 | * merge IO requests if the physical pages are ordered | |
858 | * properly. | |
859 | */ | |
535131e6 MG |
860 | list_add(&page->lru, list); |
861 | set_page_private(page, migratetype); | |
81eabcbe | 862 | list = &page->lru; |
1da177e4 | 863 | } |
c54ad30c | 864 | spin_unlock(&zone->lock); |
085cc7d5 | 865 | return i; |
1da177e4 LT |
866 | } |
867 | ||
4ae7c039 | 868 | #ifdef CONFIG_NUMA |
8fce4d8e | 869 | /* |
4037d452 CL |
870 | * Called from the vmstat counter updater to drain pagesets of this |
871 | * currently executing processor on remote nodes after they have | |
872 | * expired. | |
873 | * | |
879336c3 CL |
874 | * Note that this function must be called with the thread pinned to |
875 | * a single processor. | |
8fce4d8e | 876 | */ |
4037d452 | 877 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
4ae7c039 | 878 | { |
4ae7c039 | 879 | unsigned long flags; |
4037d452 | 880 | int to_drain; |
4ae7c039 | 881 | |
4037d452 CL |
882 | local_irq_save(flags); |
883 | if (pcp->count >= pcp->batch) | |
884 | to_drain = pcp->batch; | |
885 | else | |
886 | to_drain = pcp->count; | |
887 | free_pages_bulk(zone, to_drain, &pcp->list, 0); | |
888 | pcp->count -= to_drain; | |
889 | local_irq_restore(flags); | |
4ae7c039 CL |
890 | } |
891 | #endif | |
892 | ||
9f8f2172 CL |
893 | /* |
894 | * Drain pages of the indicated processor. | |
895 | * | |
896 | * The processor must either be the current processor and the | |
897 | * thread pinned to the current processor or a processor that | |
898 | * is not online. | |
899 | */ | |
900 | static void drain_pages(unsigned int cpu) | |
1da177e4 | 901 | { |
c54ad30c | 902 | unsigned long flags; |
1da177e4 | 903 | struct zone *zone; |
1da177e4 LT |
904 | |
905 | for_each_zone(zone) { | |
906 | struct per_cpu_pageset *pset; | |
3dfa5721 | 907 | struct per_cpu_pages *pcp; |
1da177e4 | 908 | |
f2e12bb2 CL |
909 | if (!populated_zone(zone)) |
910 | continue; | |
911 | ||
e7c8d5c9 | 912 | pset = zone_pcp(zone, cpu); |
3dfa5721 CL |
913 | |
914 | pcp = &pset->pcp; | |
915 | local_irq_save(flags); | |
916 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); | |
917 | pcp->count = 0; | |
918 | local_irq_restore(flags); | |
1da177e4 LT |
919 | } |
920 | } | |
1da177e4 | 921 | |
9f8f2172 CL |
922 | /* |
923 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
924 | */ | |
925 | void drain_local_pages(void *arg) | |
926 | { | |
927 | drain_pages(smp_processor_id()); | |
928 | } | |
929 | ||
930 | /* | |
931 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator | |
932 | */ | |
933 | void drain_all_pages(void) | |
934 | { | |
935 | on_each_cpu(drain_local_pages, NULL, 0, 1); | |
936 | } | |
937 | ||
296699de | 938 | #ifdef CONFIG_HIBERNATION |
1da177e4 LT |
939 | |
940 | void mark_free_pages(struct zone *zone) | |
941 | { | |
f623f0db RW |
942 | unsigned long pfn, max_zone_pfn; |
943 | unsigned long flags; | |
b2a0ac88 | 944 | int order, t; |
1da177e4 LT |
945 | struct list_head *curr; |
946 | ||
947 | if (!zone->spanned_pages) | |
948 | return; | |
949 | ||
950 | spin_lock_irqsave(&zone->lock, flags); | |
f623f0db RW |
951 | |
952 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
953 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
954 | if (pfn_valid(pfn)) { | |
955 | struct page *page = pfn_to_page(pfn); | |
956 | ||
7be98234 RW |
957 | if (!swsusp_page_is_forbidden(page)) |
958 | swsusp_unset_page_free(page); | |
f623f0db | 959 | } |
1da177e4 | 960 | |
b2a0ac88 MG |
961 | for_each_migratetype_order(order, t) { |
962 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
f623f0db | 963 | unsigned long i; |
1da177e4 | 964 | |
f623f0db RW |
965 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
966 | for (i = 0; i < (1UL << order); i++) | |
7be98234 | 967 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
f623f0db | 968 | } |
b2a0ac88 | 969 | } |
1da177e4 LT |
970 | spin_unlock_irqrestore(&zone->lock, flags); |
971 | } | |
e2c55dc8 | 972 | #endif /* CONFIG_PM */ |
1da177e4 | 973 | |
1da177e4 LT |
974 | /* |
975 | * Free a 0-order page | |
976 | */ | |
920c7a5d | 977 | static void free_hot_cold_page(struct page *page, int cold) |
1da177e4 LT |
978 | { |
979 | struct zone *zone = page_zone(page); | |
980 | struct per_cpu_pages *pcp; | |
981 | unsigned long flags; | |
982 | ||
1da177e4 LT |
983 | if (PageAnon(page)) |
984 | page->mapping = NULL; | |
224abf92 | 985 | if (free_pages_check(page)) |
689bcebf HD |
986 | return; |
987 | ||
9858db50 NP |
988 | if (!PageHighMem(page)) |
989 | debug_check_no_locks_freed(page_address(page), PAGE_SIZE); | |
dafb1367 | 990 | arch_free_page(page, 0); |
689bcebf HD |
991 | kernel_map_pages(page, 1, 0); |
992 | ||
3dfa5721 | 993 | pcp = &zone_pcp(zone, get_cpu())->pcp; |
1da177e4 | 994 | local_irq_save(flags); |
f8891e5e | 995 | __count_vm_event(PGFREE); |
3dfa5721 CL |
996 | if (cold) |
997 | list_add_tail(&page->lru, &pcp->list); | |
998 | else | |
999 | list_add(&page->lru, &pcp->list); | |
535131e6 | 1000 | set_page_private(page, get_pageblock_migratetype(page)); |
1da177e4 | 1001 | pcp->count++; |
48db57f8 NP |
1002 | if (pcp->count >= pcp->high) { |
1003 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
1004 | pcp->count -= pcp->batch; | |
1005 | } | |
1da177e4 LT |
1006 | local_irq_restore(flags); |
1007 | put_cpu(); | |
1008 | } | |
1009 | ||
920c7a5d | 1010 | void free_hot_page(struct page *page) |
1da177e4 LT |
1011 | { |
1012 | free_hot_cold_page(page, 0); | |
1013 | } | |
1014 | ||
920c7a5d | 1015 | void free_cold_page(struct page *page) |
1da177e4 LT |
1016 | { |
1017 | free_hot_cold_page(page, 1); | |
1018 | } | |
1019 | ||
8dfcc9ba NP |
1020 | /* |
1021 | * split_page takes a non-compound higher-order page, and splits it into | |
1022 | * n (1<<order) sub-pages: page[0..n] | |
1023 | * Each sub-page must be freed individually. | |
1024 | * | |
1025 | * Note: this is probably too low level an operation for use in drivers. | |
1026 | * Please consult with lkml before using this in your driver. | |
1027 | */ | |
1028 | void split_page(struct page *page, unsigned int order) | |
1029 | { | |
1030 | int i; | |
1031 | ||
725d704e NP |
1032 | VM_BUG_ON(PageCompound(page)); |
1033 | VM_BUG_ON(!page_count(page)); | |
7835e98b NP |
1034 | for (i = 1; i < (1 << order); i++) |
1035 | set_page_refcounted(page + i); | |
8dfcc9ba | 1036 | } |
8dfcc9ba | 1037 | |
1da177e4 LT |
1038 | /* |
1039 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1040 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1041 | * or two. | |
1042 | */ | |
a74609fa NP |
1043 | static struct page *buffered_rmqueue(struct zonelist *zonelist, |
1044 | struct zone *zone, int order, gfp_t gfp_flags) | |
1da177e4 LT |
1045 | { |
1046 | unsigned long flags; | |
689bcebf | 1047 | struct page *page; |
1da177e4 | 1048 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 1049 | int cpu; |
64c5e135 | 1050 | int migratetype = allocflags_to_migratetype(gfp_flags); |
1da177e4 | 1051 | |
689bcebf | 1052 | again: |
a74609fa | 1053 | cpu = get_cpu(); |
48db57f8 | 1054 | if (likely(order == 0)) { |
1da177e4 LT |
1055 | struct per_cpu_pages *pcp; |
1056 | ||
3dfa5721 | 1057 | pcp = &zone_pcp(zone, cpu)->pcp; |
1da177e4 | 1058 | local_irq_save(flags); |
a74609fa | 1059 | if (!pcp->count) { |
941c7105 | 1060 | pcp->count = rmqueue_bulk(zone, 0, |
b2a0ac88 | 1061 | pcp->batch, &pcp->list, migratetype); |
a74609fa NP |
1062 | if (unlikely(!pcp->count)) |
1063 | goto failed; | |
1da177e4 | 1064 | } |
b92a6edd | 1065 | |
535131e6 | 1066 | /* Find a page of the appropriate migrate type */ |
3dfa5721 CL |
1067 | if (cold) { |
1068 | list_for_each_entry_reverse(page, &pcp->list, lru) | |
1069 | if (page_private(page) == migratetype) | |
1070 | break; | |
1071 | } else { | |
1072 | list_for_each_entry(page, &pcp->list, lru) | |
1073 | if (page_private(page) == migratetype) | |
1074 | break; | |
1075 | } | |
535131e6 | 1076 | |
b92a6edd MG |
1077 | /* Allocate more to the pcp list if necessary */ |
1078 | if (unlikely(&page->lru == &pcp->list)) { | |
535131e6 MG |
1079 | pcp->count += rmqueue_bulk(zone, 0, |
1080 | pcp->batch, &pcp->list, migratetype); | |
1081 | page = list_entry(pcp->list.next, struct page, lru); | |
535131e6 | 1082 | } |
b92a6edd MG |
1083 | |
1084 | list_del(&page->lru); | |
1085 | pcp->count--; | |
7fb1d9fc | 1086 | } else { |
1da177e4 | 1087 | spin_lock_irqsave(&zone->lock, flags); |
b2a0ac88 | 1088 | page = __rmqueue(zone, order, migratetype); |
a74609fa NP |
1089 | spin_unlock(&zone->lock); |
1090 | if (!page) | |
1091 | goto failed; | |
1da177e4 LT |
1092 | } |
1093 | ||
f8891e5e | 1094 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
ca889e6c | 1095 | zone_statistics(zonelist, zone); |
a74609fa NP |
1096 | local_irq_restore(flags); |
1097 | put_cpu(); | |
1da177e4 | 1098 | |
725d704e | 1099 | VM_BUG_ON(bad_range(zone, page)); |
17cf4406 | 1100 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 1101 | goto again; |
1da177e4 | 1102 | return page; |
a74609fa NP |
1103 | |
1104 | failed: | |
1105 | local_irq_restore(flags); | |
1106 | put_cpu(); | |
1107 | return NULL; | |
1da177e4 LT |
1108 | } |
1109 | ||
7fb1d9fc | 1110 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
1111 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
1112 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
1113 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
1114 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
1115 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
1116 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 1117 | |
933e312e AM |
1118 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
1119 | ||
1120 | static struct fail_page_alloc_attr { | |
1121 | struct fault_attr attr; | |
1122 | ||
1123 | u32 ignore_gfp_highmem; | |
1124 | u32 ignore_gfp_wait; | |
54114994 | 1125 | u32 min_order; |
933e312e AM |
1126 | |
1127 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1128 | ||
1129 | struct dentry *ignore_gfp_highmem_file; | |
1130 | struct dentry *ignore_gfp_wait_file; | |
54114994 | 1131 | struct dentry *min_order_file; |
933e312e AM |
1132 | |
1133 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1134 | ||
1135 | } fail_page_alloc = { | |
1136 | .attr = FAULT_ATTR_INITIALIZER, | |
6b1b60f4 DM |
1137 | .ignore_gfp_wait = 1, |
1138 | .ignore_gfp_highmem = 1, | |
54114994 | 1139 | .min_order = 1, |
933e312e AM |
1140 | }; |
1141 | ||
1142 | static int __init setup_fail_page_alloc(char *str) | |
1143 | { | |
1144 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1145 | } | |
1146 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1147 | ||
1148 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1149 | { | |
54114994 AM |
1150 | if (order < fail_page_alloc.min_order) |
1151 | return 0; | |
933e312e AM |
1152 | if (gfp_mask & __GFP_NOFAIL) |
1153 | return 0; | |
1154 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
1155 | return 0; | |
1156 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
1157 | return 0; | |
1158 | ||
1159 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1160 | } | |
1161 | ||
1162 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1163 | ||
1164 | static int __init fail_page_alloc_debugfs(void) | |
1165 | { | |
1166 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
1167 | struct dentry *dir; | |
1168 | int err; | |
1169 | ||
1170 | err = init_fault_attr_dentries(&fail_page_alloc.attr, | |
1171 | "fail_page_alloc"); | |
1172 | if (err) | |
1173 | return err; | |
1174 | dir = fail_page_alloc.attr.dentries.dir; | |
1175 | ||
1176 | fail_page_alloc.ignore_gfp_wait_file = | |
1177 | debugfs_create_bool("ignore-gfp-wait", mode, dir, | |
1178 | &fail_page_alloc.ignore_gfp_wait); | |
1179 | ||
1180 | fail_page_alloc.ignore_gfp_highmem_file = | |
1181 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1182 | &fail_page_alloc.ignore_gfp_highmem); | |
54114994 AM |
1183 | fail_page_alloc.min_order_file = |
1184 | debugfs_create_u32("min-order", mode, dir, | |
1185 | &fail_page_alloc.min_order); | |
933e312e AM |
1186 | |
1187 | if (!fail_page_alloc.ignore_gfp_wait_file || | |
54114994 AM |
1188 | !fail_page_alloc.ignore_gfp_highmem_file || |
1189 | !fail_page_alloc.min_order_file) { | |
933e312e AM |
1190 | err = -ENOMEM; |
1191 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); | |
1192 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); | |
54114994 | 1193 | debugfs_remove(fail_page_alloc.min_order_file); |
933e312e AM |
1194 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); |
1195 | } | |
1196 | ||
1197 | return err; | |
1198 | } | |
1199 | ||
1200 | late_initcall(fail_page_alloc_debugfs); | |
1201 | ||
1202 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1203 | ||
1204 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1205 | ||
1206 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1207 | { | |
1208 | return 0; | |
1209 | } | |
1210 | ||
1211 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1212 | ||
1da177e4 LT |
1213 | /* |
1214 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
1215 | * of the allocation. | |
1216 | */ | |
1217 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 1218 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
1219 | { |
1220 | /* free_pages my go negative - that's OK */ | |
d23ad423 CL |
1221 | long min = mark; |
1222 | long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1; | |
1da177e4 LT |
1223 | int o; |
1224 | ||
7fb1d9fc | 1225 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 1226 | min -= min / 2; |
7fb1d9fc | 1227 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
1228 | min -= min / 4; |
1229 | ||
1230 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
1231 | return 0; | |
1232 | for (o = 0; o < order; o++) { | |
1233 | /* At the next order, this order's pages become unavailable */ | |
1234 | free_pages -= z->free_area[o].nr_free << o; | |
1235 | ||
1236 | /* Require fewer higher order pages to be free */ | |
1237 | min >>= 1; | |
1238 | ||
1239 | if (free_pages <= min) | |
1240 | return 0; | |
1241 | } | |
1242 | return 1; | |
1243 | } | |
1244 | ||
9276b1bc PJ |
1245 | #ifdef CONFIG_NUMA |
1246 | /* | |
1247 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1248 | * skip over zones that are not allowed by the cpuset, or that have | |
1249 | * been recently (in last second) found to be nearly full. See further | |
1250 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
183ff22b | 1251 | * that have to skip over a lot of full or unallowed zones. |
9276b1bc PJ |
1252 | * |
1253 | * If the zonelist cache is present in the passed in zonelist, then | |
1254 | * returns a pointer to the allowed node mask (either the current | |
37b07e41 | 1255 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) |
9276b1bc PJ |
1256 | * |
1257 | * If the zonelist cache is not available for this zonelist, does | |
1258 | * nothing and returns NULL. | |
1259 | * | |
1260 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1261 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1262 | * | |
1263 | * We hold off even calling zlc_setup, until after we've checked the | |
1264 | * first zone in the zonelist, on the theory that most allocations will | |
1265 | * be satisfied from that first zone, so best to examine that zone as | |
1266 | * quickly as we can. | |
1267 | */ | |
1268 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1269 | { | |
1270 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1271 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1272 | ||
1273 | zlc = zonelist->zlcache_ptr; | |
1274 | if (!zlc) | |
1275 | return NULL; | |
1276 | ||
1277 | if (jiffies - zlc->last_full_zap > 1 * HZ) { | |
1278 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1279 | zlc->last_full_zap = jiffies; | |
1280 | } | |
1281 | ||
1282 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1283 | &cpuset_current_mems_allowed : | |
37b07e41 | 1284 | &node_states[N_HIGH_MEMORY]; |
9276b1bc PJ |
1285 | return allowednodes; |
1286 | } | |
1287 | ||
1288 | /* | |
1289 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1290 | * if it is worth looking at further for free memory: | |
1291 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1292 | * bit set in the zonelist_cache fullzones BITMAP). | |
1293 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1294 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1295 | * Return true (non-zero) if zone is worth looking at further, or | |
1296 | * else return false (zero) if it is not. | |
1297 | * | |
1298 | * This check -ignores- the distinction between various watermarks, | |
1299 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1300 | * found to be full for any variation of these watermarks, it will | |
1301 | * be considered full for up to one second by all requests, unless | |
1302 | * we are so low on memory on all allowed nodes that we are forced | |
1303 | * into the second scan of the zonelist. | |
1304 | * | |
1305 | * In the second scan we ignore this zonelist cache and exactly | |
1306 | * apply the watermarks to all zones, even it is slower to do so. | |
1307 | * We are low on memory in the second scan, and should leave no stone | |
1308 | * unturned looking for a free page. | |
1309 | */ | |
1310 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z, | |
1311 | nodemask_t *allowednodes) | |
1312 | { | |
1313 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1314 | int i; /* index of *z in zonelist zones */ | |
1315 | int n; /* node that zone *z is on */ | |
1316 | ||
1317 | zlc = zonelist->zlcache_ptr; | |
1318 | if (!zlc) | |
1319 | return 1; | |
1320 | ||
1321 | i = z - zonelist->zones; | |
1322 | n = zlc->z_to_n[i]; | |
1323 | ||
1324 | /* This zone is worth trying if it is allowed but not full */ | |
1325 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1326 | } | |
1327 | ||
1328 | /* | |
1329 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1330 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1331 | * from that zone don't waste time re-examining it. | |
1332 | */ | |
1333 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z) | |
1334 | { | |
1335 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1336 | int i; /* index of *z in zonelist zones */ | |
1337 | ||
1338 | zlc = zonelist->zlcache_ptr; | |
1339 | if (!zlc) | |
1340 | return; | |
1341 | ||
1342 | i = z - zonelist->zones; | |
1343 | ||
1344 | set_bit(i, zlc->fullzones); | |
1345 | } | |
1346 | ||
1347 | #else /* CONFIG_NUMA */ | |
1348 | ||
1349 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1350 | { | |
1351 | return NULL; | |
1352 | } | |
1353 | ||
1354 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z, | |
1355 | nodemask_t *allowednodes) | |
1356 | { | |
1357 | return 1; | |
1358 | } | |
1359 | ||
1360 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z) | |
1361 | { | |
1362 | } | |
1363 | #endif /* CONFIG_NUMA */ | |
1364 | ||
7fb1d9fc | 1365 | /* |
0798e519 | 1366 | * get_page_from_freelist goes through the zonelist trying to allocate |
7fb1d9fc RS |
1367 | * a page. |
1368 | */ | |
1369 | static struct page * | |
1370 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, | |
1371 | struct zonelist *zonelist, int alloc_flags) | |
753ee728 | 1372 | { |
9276b1bc | 1373 | struct zone **z; |
7fb1d9fc | 1374 | struct page *page = NULL; |
9276b1bc | 1375 | int classzone_idx = zone_idx(zonelist->zones[0]); |
1192d526 | 1376 | struct zone *zone; |
9276b1bc PJ |
1377 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
1378 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1379 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
b377fd39 | 1380 | enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */ |
7fb1d9fc | 1381 | |
9276b1bc | 1382 | zonelist_scan: |
7fb1d9fc | 1383 | /* |
9276b1bc | 1384 | * Scan zonelist, looking for a zone with enough free. |
7fb1d9fc RS |
1385 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1386 | */ | |
9276b1bc PJ |
1387 | z = zonelist->zones; |
1388 | ||
7fb1d9fc | 1389 | do { |
b377fd39 MG |
1390 | /* |
1391 | * In NUMA, this could be a policy zonelist which contains | |
1392 | * zones that may not be allowed by the current gfp_mask. | |
1393 | * Check the zone is allowed by the current flags | |
1394 | */ | |
1395 | if (unlikely(alloc_should_filter_zonelist(zonelist))) { | |
1396 | if (highest_zoneidx == -1) | |
1397 | highest_zoneidx = gfp_zone(gfp_mask); | |
1398 | if (zone_idx(*z) > highest_zoneidx) | |
1399 | continue; | |
1400 | } | |
1401 | ||
9276b1bc PJ |
1402 | if (NUMA_BUILD && zlc_active && |
1403 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1404 | continue; | |
1192d526 | 1405 | zone = *z; |
7fb1d9fc | 1406 | if ((alloc_flags & ALLOC_CPUSET) && |
02a0e53d | 1407 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
9276b1bc | 1408 | goto try_next_zone; |
7fb1d9fc RS |
1409 | |
1410 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
1411 | unsigned long mark; |
1412 | if (alloc_flags & ALLOC_WMARK_MIN) | |
1192d526 | 1413 | mark = zone->pages_min; |
3148890b | 1414 | else if (alloc_flags & ALLOC_WMARK_LOW) |
1192d526 | 1415 | mark = zone->pages_low; |
3148890b | 1416 | else |
1192d526 | 1417 | mark = zone->pages_high; |
0798e519 PJ |
1418 | if (!zone_watermark_ok(zone, order, mark, |
1419 | classzone_idx, alloc_flags)) { | |
9eeff239 | 1420 | if (!zone_reclaim_mode || |
1192d526 | 1421 | !zone_reclaim(zone, gfp_mask, order)) |
9276b1bc | 1422 | goto this_zone_full; |
0798e519 | 1423 | } |
7fb1d9fc RS |
1424 | } |
1425 | ||
1192d526 | 1426 | page = buffered_rmqueue(zonelist, zone, order, gfp_mask); |
0798e519 | 1427 | if (page) |
7fb1d9fc | 1428 | break; |
9276b1bc PJ |
1429 | this_zone_full: |
1430 | if (NUMA_BUILD) | |
1431 | zlc_mark_zone_full(zonelist, z); | |
1432 | try_next_zone: | |
1433 | if (NUMA_BUILD && !did_zlc_setup) { | |
1434 | /* we do zlc_setup after the first zone is tried */ | |
1435 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1436 | zlc_active = 1; | |
1437 | did_zlc_setup = 1; | |
1438 | } | |
7fb1d9fc | 1439 | } while (*(++z) != NULL); |
9276b1bc PJ |
1440 | |
1441 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1442 | /* Disable zlc cache for second zonelist scan */ | |
1443 | zlc_active = 0; | |
1444 | goto zonelist_scan; | |
1445 | } | |
7fb1d9fc | 1446 | return page; |
753ee728 MH |
1447 | } |
1448 | ||
1da177e4 LT |
1449 | /* |
1450 | * This is the 'heart' of the zoned buddy allocator. | |
1451 | */ | |
1452 | struct page * fastcall | |
dd0fc66f | 1453 | __alloc_pages(gfp_t gfp_mask, unsigned int order, |
1da177e4 LT |
1454 | struct zonelist *zonelist) |
1455 | { | |
260b2367 | 1456 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
7fb1d9fc | 1457 | struct zone **z; |
1da177e4 LT |
1458 | struct page *page; |
1459 | struct reclaim_state reclaim_state; | |
1460 | struct task_struct *p = current; | |
1da177e4 | 1461 | int do_retry; |
7fb1d9fc | 1462 | int alloc_flags; |
1da177e4 LT |
1463 | int did_some_progress; |
1464 | ||
1465 | might_sleep_if(wait); | |
1466 | ||
933e312e AM |
1467 | if (should_fail_alloc_page(gfp_mask, order)) |
1468 | return NULL; | |
1469 | ||
6b1de916 | 1470 | restart: |
7fb1d9fc | 1471 | z = zonelist->zones; /* the list of zones suitable for gfp_mask */ |
1da177e4 | 1472 | |
7fb1d9fc | 1473 | if (unlikely(*z == NULL)) { |
523b9458 CL |
1474 | /* |
1475 | * Happens if we have an empty zonelist as a result of | |
1476 | * GFP_THISNODE being used on a memoryless node | |
1477 | */ | |
1da177e4 LT |
1478 | return NULL; |
1479 | } | |
6b1de916 | 1480 | |
7fb1d9fc | 1481 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1482 | zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET); |
7fb1d9fc RS |
1483 | if (page) |
1484 | goto got_pg; | |
1da177e4 | 1485 | |
952f3b51 CL |
1486 | /* |
1487 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
1488 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
1489 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
1490 | * using a larger set of nodes after it has established that the | |
1491 | * allowed per node queues are empty and that nodes are | |
1492 | * over allocated. | |
1493 | */ | |
1494 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
1495 | goto nopage; | |
1496 | ||
0798e519 | 1497 | for (z = zonelist->zones; *z; z++) |
43b0bc00 | 1498 | wakeup_kswapd(*z, order); |
1da177e4 | 1499 | |
9bf2229f | 1500 | /* |
7fb1d9fc RS |
1501 | * OK, we're below the kswapd watermark and have kicked background |
1502 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1503 | * to how we want to proceed. | |
1504 | * | |
1505 | * The caller may dip into page reserves a bit more if the caller | |
1506 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
1507 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
1508 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 1509 | */ |
3148890b | 1510 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
1511 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
1512 | alloc_flags |= ALLOC_HARDER; | |
1513 | if (gfp_mask & __GFP_HIGH) | |
1514 | alloc_flags |= ALLOC_HIGH; | |
bdd804f4 PJ |
1515 | if (wait) |
1516 | alloc_flags |= ALLOC_CPUSET; | |
1da177e4 LT |
1517 | |
1518 | /* | |
1519 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 1520 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
1521 | * |
1522 | * This is the last chance, in general, before the goto nopage. | |
1523 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 1524 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 1525 | */ |
7fb1d9fc RS |
1526 | page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags); |
1527 | if (page) | |
1528 | goto got_pg; | |
1da177e4 LT |
1529 | |
1530 | /* This allocation should allow future memory freeing. */ | |
b84a35be | 1531 | |
b43a57bb | 1532 | rebalance: |
b84a35be NP |
1533 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) |
1534 | && !in_interrupt()) { | |
1535 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
885036d3 | 1536 | nofail_alloc: |
b84a35be | 1537 | /* go through the zonelist yet again, ignoring mins */ |
7fb1d9fc | 1538 | page = get_page_from_freelist(gfp_mask, order, |
47f3a867 | 1539 | zonelist, ALLOC_NO_WATERMARKS); |
7fb1d9fc RS |
1540 | if (page) |
1541 | goto got_pg; | |
885036d3 | 1542 | if (gfp_mask & __GFP_NOFAIL) { |
3fcfab16 | 1543 | congestion_wait(WRITE, HZ/50); |
885036d3 KK |
1544 | goto nofail_alloc; |
1545 | } | |
1da177e4 LT |
1546 | } |
1547 | goto nopage; | |
1548 | } | |
1549 | ||
1550 | /* Atomic allocations - we can't balance anything */ | |
1551 | if (!wait) | |
1552 | goto nopage; | |
1553 | ||
1da177e4 LT |
1554 | cond_resched(); |
1555 | ||
1556 | /* We now go into synchronous reclaim */ | |
3e0d98b9 | 1557 | cpuset_memory_pressure_bump(); |
1da177e4 LT |
1558 | p->flags |= PF_MEMALLOC; |
1559 | reclaim_state.reclaimed_slab = 0; | |
1560 | p->reclaim_state = &reclaim_state; | |
1561 | ||
5ad333eb | 1562 | did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask); |
1da177e4 LT |
1563 | |
1564 | p->reclaim_state = NULL; | |
1565 | p->flags &= ~PF_MEMALLOC; | |
1566 | ||
1567 | cond_resched(); | |
1568 | ||
e2c55dc8 | 1569 | if (order != 0) |
9f8f2172 | 1570 | drain_all_pages(); |
e2c55dc8 | 1571 | |
1da177e4 | 1572 | if (likely(did_some_progress)) { |
7fb1d9fc RS |
1573 | page = get_page_from_freelist(gfp_mask, order, |
1574 | zonelist, alloc_flags); | |
1575 | if (page) | |
1576 | goto got_pg; | |
1da177e4 | 1577 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
ff0ceb9d DR |
1578 | if (!try_set_zone_oom(zonelist)) { |
1579 | schedule_timeout_uninterruptible(1); | |
1580 | goto restart; | |
1581 | } | |
1582 | ||
1da177e4 LT |
1583 | /* |
1584 | * Go through the zonelist yet one more time, keep | |
1585 | * very high watermark here, this is only to catch | |
1586 | * a parallel oom killing, we must fail if we're still | |
1587 | * under heavy pressure. | |
1588 | */ | |
7fb1d9fc | 1589 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1590 | zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET); |
ff0ceb9d DR |
1591 | if (page) { |
1592 | clear_zonelist_oom(zonelist); | |
7fb1d9fc | 1593 | goto got_pg; |
ff0ceb9d | 1594 | } |
1da177e4 | 1595 | |
a8bbf72a | 1596 | /* The OOM killer will not help higher order allocs so fail */ |
ff0ceb9d DR |
1597 | if (order > PAGE_ALLOC_COSTLY_ORDER) { |
1598 | clear_zonelist_oom(zonelist); | |
a8bbf72a | 1599 | goto nopage; |
ff0ceb9d | 1600 | } |
a8bbf72a | 1601 | |
9b0f8b04 | 1602 | out_of_memory(zonelist, gfp_mask, order); |
ff0ceb9d | 1603 | clear_zonelist_oom(zonelist); |
1da177e4 LT |
1604 | goto restart; |
1605 | } | |
1606 | ||
1607 | /* | |
1608 | * Don't let big-order allocations loop unless the caller explicitly | |
1609 | * requests that. Wait for some write requests to complete then retry. | |
1610 | * | |
1611 | * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order | |
1612 | * <= 3, but that may not be true in other implementations. | |
1613 | */ | |
1614 | do_retry = 0; | |
1615 | if (!(gfp_mask & __GFP_NORETRY)) { | |
5ad333eb AW |
1616 | if ((order <= PAGE_ALLOC_COSTLY_ORDER) || |
1617 | (gfp_mask & __GFP_REPEAT)) | |
1da177e4 LT |
1618 | do_retry = 1; |
1619 | if (gfp_mask & __GFP_NOFAIL) | |
1620 | do_retry = 1; | |
1621 | } | |
1622 | if (do_retry) { | |
3fcfab16 | 1623 | congestion_wait(WRITE, HZ/50); |
1da177e4 LT |
1624 | goto rebalance; |
1625 | } | |
1626 | ||
1627 | nopage: | |
1628 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1629 | printk(KERN_WARNING "%s: page allocation failure." | |
1630 | " order:%d, mode:0x%x\n", | |
1631 | p->comm, order, gfp_mask); | |
1632 | dump_stack(); | |
578c2fd6 | 1633 | show_mem(); |
1da177e4 | 1634 | } |
1da177e4 | 1635 | got_pg: |
1da177e4 LT |
1636 | return page; |
1637 | } | |
1638 | ||
1639 | EXPORT_SYMBOL(__alloc_pages); | |
1640 | ||
1641 | /* | |
1642 | * Common helper functions. | |
1643 | */ | |
920c7a5d | 1644 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1645 | { |
1646 | struct page * page; | |
1647 | page = alloc_pages(gfp_mask, order); | |
1648 | if (!page) | |
1649 | return 0; | |
1650 | return (unsigned long) page_address(page); | |
1651 | } | |
1652 | ||
1653 | EXPORT_SYMBOL(__get_free_pages); | |
1654 | ||
920c7a5d | 1655 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1656 | { |
1657 | struct page * page; | |
1658 | ||
1659 | /* | |
1660 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1661 | * a highmem page | |
1662 | */ | |
725d704e | 1663 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1664 | |
1665 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1666 | if (page) | |
1667 | return (unsigned long) page_address(page); | |
1668 | return 0; | |
1669 | } | |
1670 | ||
1671 | EXPORT_SYMBOL(get_zeroed_page); | |
1672 | ||
1673 | void __pagevec_free(struct pagevec *pvec) | |
1674 | { | |
1675 | int i = pagevec_count(pvec); | |
1676 | ||
1677 | while (--i >= 0) | |
1678 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1679 | } | |
1680 | ||
920c7a5d | 1681 | void __free_pages(struct page *page, unsigned int order) |
1da177e4 | 1682 | { |
b5810039 | 1683 | if (put_page_testzero(page)) { |
1da177e4 LT |
1684 | if (order == 0) |
1685 | free_hot_page(page); | |
1686 | else | |
1687 | __free_pages_ok(page, order); | |
1688 | } | |
1689 | } | |
1690 | ||
1691 | EXPORT_SYMBOL(__free_pages); | |
1692 | ||
920c7a5d | 1693 | void free_pages(unsigned long addr, unsigned int order) |
1da177e4 LT |
1694 | { |
1695 | if (addr != 0) { | |
725d704e | 1696 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
1da177e4 LT |
1697 | __free_pages(virt_to_page((void *)addr), order); |
1698 | } | |
1699 | } | |
1700 | ||
1701 | EXPORT_SYMBOL(free_pages); | |
1702 | ||
1da177e4 LT |
1703 | static unsigned int nr_free_zone_pages(int offset) |
1704 | { | |
e310fd43 MB |
1705 | /* Just pick one node, since fallback list is circular */ |
1706 | pg_data_t *pgdat = NODE_DATA(numa_node_id()); | |
1da177e4 LT |
1707 | unsigned int sum = 0; |
1708 | ||
e310fd43 MB |
1709 | struct zonelist *zonelist = pgdat->node_zonelists + offset; |
1710 | struct zone **zonep = zonelist->zones; | |
1711 | struct zone *zone; | |
1da177e4 | 1712 | |
e310fd43 MB |
1713 | for (zone = *zonep++; zone; zone = *zonep++) { |
1714 | unsigned long size = zone->present_pages; | |
1715 | unsigned long high = zone->pages_high; | |
1716 | if (size > high) | |
1717 | sum += size - high; | |
1da177e4 LT |
1718 | } |
1719 | ||
1720 | return sum; | |
1721 | } | |
1722 | ||
1723 | /* | |
1724 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1725 | */ | |
1726 | unsigned int nr_free_buffer_pages(void) | |
1727 | { | |
af4ca457 | 1728 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 | 1729 | } |
c2f1a551 | 1730 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
1da177e4 LT |
1731 | |
1732 | /* | |
1733 | * Amount of free RAM allocatable within all zones | |
1734 | */ | |
1735 | unsigned int nr_free_pagecache_pages(void) | |
1736 | { | |
2a1e274a | 1737 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
1da177e4 | 1738 | } |
08e0f6a9 CL |
1739 | |
1740 | static inline void show_node(struct zone *zone) | |
1da177e4 | 1741 | { |
08e0f6a9 | 1742 | if (NUMA_BUILD) |
25ba77c1 | 1743 | printk("Node %d ", zone_to_nid(zone)); |
1da177e4 | 1744 | } |
1da177e4 | 1745 | |
1da177e4 LT |
1746 | void si_meminfo(struct sysinfo *val) |
1747 | { | |
1748 | val->totalram = totalram_pages; | |
1749 | val->sharedram = 0; | |
d23ad423 | 1750 | val->freeram = global_page_state(NR_FREE_PAGES); |
1da177e4 | 1751 | val->bufferram = nr_blockdev_pages(); |
1da177e4 LT |
1752 | val->totalhigh = totalhigh_pages; |
1753 | val->freehigh = nr_free_highpages(); | |
1da177e4 LT |
1754 | val->mem_unit = PAGE_SIZE; |
1755 | } | |
1756 | ||
1757 | EXPORT_SYMBOL(si_meminfo); | |
1758 | ||
1759 | #ifdef CONFIG_NUMA | |
1760 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1761 | { | |
1762 | pg_data_t *pgdat = NODE_DATA(nid); | |
1763 | ||
1764 | val->totalram = pgdat->node_present_pages; | |
d23ad423 | 1765 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
98d2b0eb | 1766 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 1767 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
d23ad423 CL |
1768 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
1769 | NR_FREE_PAGES); | |
98d2b0eb CL |
1770 | #else |
1771 | val->totalhigh = 0; | |
1772 | val->freehigh = 0; | |
1773 | #endif | |
1da177e4 LT |
1774 | val->mem_unit = PAGE_SIZE; |
1775 | } | |
1776 | #endif | |
1777 | ||
1778 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1779 | ||
1780 | /* | |
1781 | * Show free area list (used inside shift_scroll-lock stuff) | |
1782 | * We also calculate the percentage fragmentation. We do this by counting the | |
1783 | * memory on each free list with the exception of the first item on the list. | |
1784 | */ | |
1785 | void show_free_areas(void) | |
1786 | { | |
c7241913 | 1787 | int cpu; |
1da177e4 LT |
1788 | struct zone *zone; |
1789 | ||
1790 | for_each_zone(zone) { | |
c7241913 | 1791 | if (!populated_zone(zone)) |
1da177e4 | 1792 | continue; |
c7241913 JS |
1793 | |
1794 | show_node(zone); | |
1795 | printk("%s per-cpu:\n", zone->name); | |
1da177e4 | 1796 | |
6b482c67 | 1797 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1798 | struct per_cpu_pageset *pageset; |
1799 | ||
e7c8d5c9 | 1800 | pageset = zone_pcp(zone, cpu); |
1da177e4 | 1801 | |
3dfa5721 CL |
1802 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
1803 | cpu, pageset->pcp.high, | |
1804 | pageset->pcp.batch, pageset->pcp.count); | |
1da177e4 LT |
1805 | } |
1806 | } | |
1807 | ||
a25700a5 | 1808 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n" |
d23ad423 | 1809 | " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n", |
65e458d4 CL |
1810 | global_page_state(NR_ACTIVE), |
1811 | global_page_state(NR_INACTIVE), | |
b1e7a8fd | 1812 | global_page_state(NR_FILE_DIRTY), |
ce866b34 | 1813 | global_page_state(NR_WRITEBACK), |
fd39fc85 | 1814 | global_page_state(NR_UNSTABLE_NFS), |
d23ad423 | 1815 | global_page_state(NR_FREE_PAGES), |
972d1a7b CL |
1816 | global_page_state(NR_SLAB_RECLAIMABLE) + |
1817 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
65ba55f5 | 1818 | global_page_state(NR_FILE_MAPPED), |
a25700a5 AM |
1819 | global_page_state(NR_PAGETABLE), |
1820 | global_page_state(NR_BOUNCE)); | |
1da177e4 LT |
1821 | |
1822 | for_each_zone(zone) { | |
1823 | int i; | |
1824 | ||
c7241913 JS |
1825 | if (!populated_zone(zone)) |
1826 | continue; | |
1827 | ||
1da177e4 LT |
1828 | show_node(zone); |
1829 | printk("%s" | |
1830 | " free:%lukB" | |
1831 | " min:%lukB" | |
1832 | " low:%lukB" | |
1833 | " high:%lukB" | |
1834 | " active:%lukB" | |
1835 | " inactive:%lukB" | |
1836 | " present:%lukB" | |
1837 | " pages_scanned:%lu" | |
1838 | " all_unreclaimable? %s" | |
1839 | "\n", | |
1840 | zone->name, | |
d23ad423 | 1841 | K(zone_page_state(zone, NR_FREE_PAGES)), |
1da177e4 LT |
1842 | K(zone->pages_min), |
1843 | K(zone->pages_low), | |
1844 | K(zone->pages_high), | |
c8785385 CL |
1845 | K(zone_page_state(zone, NR_ACTIVE)), |
1846 | K(zone_page_state(zone, NR_INACTIVE)), | |
1da177e4 LT |
1847 | K(zone->present_pages), |
1848 | zone->pages_scanned, | |
e815af95 | 1849 | (zone_is_all_unreclaimable(zone) ? "yes" : "no") |
1da177e4 LT |
1850 | ); |
1851 | printk("lowmem_reserve[]:"); | |
1852 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1853 | printk(" %lu", zone->lowmem_reserve[i]); | |
1854 | printk("\n"); | |
1855 | } | |
1856 | ||
1857 | for_each_zone(zone) { | |
8f9de51a | 1858 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
1da177e4 | 1859 | |
c7241913 JS |
1860 | if (!populated_zone(zone)) |
1861 | continue; | |
1862 | ||
1da177e4 LT |
1863 | show_node(zone); |
1864 | printk("%s: ", zone->name); | |
1da177e4 LT |
1865 | |
1866 | spin_lock_irqsave(&zone->lock, flags); | |
1867 | for (order = 0; order < MAX_ORDER; order++) { | |
8f9de51a KK |
1868 | nr[order] = zone->free_area[order].nr_free; |
1869 | total += nr[order] << order; | |
1da177e4 LT |
1870 | } |
1871 | spin_unlock_irqrestore(&zone->lock, flags); | |
8f9de51a KK |
1872 | for (order = 0; order < MAX_ORDER; order++) |
1873 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1da177e4 LT |
1874 | printk("= %lukB\n", K(total)); |
1875 | } | |
1876 | ||
1877 | show_swap_cache_info(); | |
1878 | } | |
1879 | ||
1880 | /* | |
1881 | * Builds allocation fallback zone lists. | |
1a93205b CL |
1882 | * |
1883 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 1884 | */ |
f0c0b2b8 KH |
1885 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
1886 | int nr_zones, enum zone_type zone_type) | |
1da177e4 | 1887 | { |
1a93205b CL |
1888 | struct zone *zone; |
1889 | ||
98d2b0eb | 1890 | BUG_ON(zone_type >= MAX_NR_ZONES); |
2f6726e5 | 1891 | zone_type++; |
02a68a5e CL |
1892 | |
1893 | do { | |
2f6726e5 | 1894 | zone_type--; |
070f8032 | 1895 | zone = pgdat->node_zones + zone_type; |
1a93205b | 1896 | if (populated_zone(zone)) { |
070f8032 CL |
1897 | zonelist->zones[nr_zones++] = zone; |
1898 | check_highest_zone(zone_type); | |
1da177e4 | 1899 | } |
02a68a5e | 1900 | |
2f6726e5 | 1901 | } while (zone_type); |
070f8032 | 1902 | return nr_zones; |
1da177e4 LT |
1903 | } |
1904 | ||
f0c0b2b8 KH |
1905 | |
1906 | /* | |
1907 | * zonelist_order: | |
1908 | * 0 = automatic detection of better ordering. | |
1909 | * 1 = order by ([node] distance, -zonetype) | |
1910 | * 2 = order by (-zonetype, [node] distance) | |
1911 | * | |
1912 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
1913 | * the same zonelist. So only NUMA can configure this param. | |
1914 | */ | |
1915 | #define ZONELIST_ORDER_DEFAULT 0 | |
1916 | #define ZONELIST_ORDER_NODE 1 | |
1917 | #define ZONELIST_ORDER_ZONE 2 | |
1918 | ||
1919 | /* zonelist order in the kernel. | |
1920 | * set_zonelist_order() will set this to NODE or ZONE. | |
1921 | */ | |
1922 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1923 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
1924 | ||
1925 | ||
1da177e4 | 1926 | #ifdef CONFIG_NUMA |
f0c0b2b8 KH |
1927 | /* The value user specified ....changed by config */ |
1928 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1929 | /* string for sysctl */ | |
1930 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
1931 | char numa_zonelist_order[16] = "default"; | |
1932 | ||
1933 | /* | |
1934 | * interface for configure zonelist ordering. | |
1935 | * command line option "numa_zonelist_order" | |
1936 | * = "[dD]efault - default, automatic configuration. | |
1937 | * = "[nN]ode - order by node locality, then by zone within node | |
1938 | * = "[zZ]one - order by zone, then by locality within zone | |
1939 | */ | |
1940 | ||
1941 | static int __parse_numa_zonelist_order(char *s) | |
1942 | { | |
1943 | if (*s == 'd' || *s == 'D') { | |
1944 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
1945 | } else if (*s == 'n' || *s == 'N') { | |
1946 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
1947 | } else if (*s == 'z' || *s == 'Z') { | |
1948 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
1949 | } else { | |
1950 | printk(KERN_WARNING | |
1951 | "Ignoring invalid numa_zonelist_order value: " | |
1952 | "%s\n", s); | |
1953 | return -EINVAL; | |
1954 | } | |
1955 | return 0; | |
1956 | } | |
1957 | ||
1958 | static __init int setup_numa_zonelist_order(char *s) | |
1959 | { | |
1960 | if (s) | |
1961 | return __parse_numa_zonelist_order(s); | |
1962 | return 0; | |
1963 | } | |
1964 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
1965 | ||
1966 | /* | |
1967 | * sysctl handler for numa_zonelist_order | |
1968 | */ | |
1969 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
1970 | struct file *file, void __user *buffer, size_t *length, | |
1971 | loff_t *ppos) | |
1972 | { | |
1973 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
1974 | int ret; | |
1975 | ||
1976 | if (write) | |
1977 | strncpy(saved_string, (char*)table->data, | |
1978 | NUMA_ZONELIST_ORDER_LEN); | |
1979 | ret = proc_dostring(table, write, file, buffer, length, ppos); | |
1980 | if (ret) | |
1981 | return ret; | |
1982 | if (write) { | |
1983 | int oldval = user_zonelist_order; | |
1984 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
1985 | /* | |
1986 | * bogus value. restore saved string | |
1987 | */ | |
1988 | strncpy((char*)table->data, saved_string, | |
1989 | NUMA_ZONELIST_ORDER_LEN); | |
1990 | user_zonelist_order = oldval; | |
1991 | } else if (oldval != user_zonelist_order) | |
1992 | build_all_zonelists(); | |
1993 | } | |
1994 | return 0; | |
1995 | } | |
1996 | ||
1997 | ||
1da177e4 | 1998 | #define MAX_NODE_LOAD (num_online_nodes()) |
f0c0b2b8 KH |
1999 | static int node_load[MAX_NUMNODES]; |
2000 | ||
1da177e4 | 2001 | /** |
4dc3b16b | 2002 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
2003 | * @node: node whose fallback list we're appending |
2004 | * @used_node_mask: nodemask_t of already used nodes | |
2005 | * | |
2006 | * We use a number of factors to determine which is the next node that should | |
2007 | * appear on a given node's fallback list. The node should not have appeared | |
2008 | * already in @node's fallback list, and it should be the next closest node | |
2009 | * according to the distance array (which contains arbitrary distance values | |
2010 | * from each node to each node in the system), and should also prefer nodes | |
2011 | * with no CPUs, since presumably they'll have very little allocation pressure | |
2012 | * on them otherwise. | |
2013 | * It returns -1 if no node is found. | |
2014 | */ | |
f0c0b2b8 | 2015 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
1da177e4 | 2016 | { |
4cf808eb | 2017 | int n, val; |
1da177e4 LT |
2018 | int min_val = INT_MAX; |
2019 | int best_node = -1; | |
2020 | ||
4cf808eb LT |
2021 | /* Use the local node if we haven't already */ |
2022 | if (!node_isset(node, *used_node_mask)) { | |
2023 | node_set(node, *used_node_mask); | |
2024 | return node; | |
2025 | } | |
1da177e4 | 2026 | |
37b07e41 | 2027 | for_each_node_state(n, N_HIGH_MEMORY) { |
4cf808eb | 2028 | cpumask_t tmp; |
1da177e4 LT |
2029 | |
2030 | /* Don't want a node to appear more than once */ | |
2031 | if (node_isset(n, *used_node_mask)) | |
2032 | continue; | |
2033 | ||
1da177e4 LT |
2034 | /* Use the distance array to find the distance */ |
2035 | val = node_distance(node, n); | |
2036 | ||
4cf808eb LT |
2037 | /* Penalize nodes under us ("prefer the next node") */ |
2038 | val += (n < node); | |
2039 | ||
1da177e4 LT |
2040 | /* Give preference to headless and unused nodes */ |
2041 | tmp = node_to_cpumask(n); | |
2042 | if (!cpus_empty(tmp)) | |
2043 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
2044 | ||
2045 | /* Slight preference for less loaded node */ | |
2046 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
2047 | val += node_load[n]; | |
2048 | ||
2049 | if (val < min_val) { | |
2050 | min_val = val; | |
2051 | best_node = n; | |
2052 | } | |
2053 | } | |
2054 | ||
2055 | if (best_node >= 0) | |
2056 | node_set(best_node, *used_node_mask); | |
2057 | ||
2058 | return best_node; | |
2059 | } | |
2060 | ||
f0c0b2b8 KH |
2061 | |
2062 | /* | |
2063 | * Build zonelists ordered by node and zones within node. | |
2064 | * This results in maximum locality--normal zone overflows into local | |
2065 | * DMA zone, if any--but risks exhausting DMA zone. | |
2066 | */ | |
2067 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
1da177e4 | 2068 | { |
19655d34 | 2069 | enum zone_type i; |
f0c0b2b8 | 2070 | int j; |
1da177e4 | 2071 | struct zonelist *zonelist; |
f0c0b2b8 KH |
2072 | |
2073 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2074 | zonelist = pgdat->node_zonelists + i; | |
2075 | for (j = 0; zonelist->zones[j] != NULL; j++) | |
2076 | ; | |
2077 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); | |
2078 | zonelist->zones[j] = NULL; | |
2079 | } | |
2080 | } | |
2081 | ||
523b9458 CL |
2082 | /* |
2083 | * Build gfp_thisnode zonelists | |
2084 | */ | |
2085 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
2086 | { | |
2087 | enum zone_type i; | |
2088 | int j; | |
2089 | struct zonelist *zonelist; | |
2090 | ||
2091 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2092 | zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i; | |
2093 | j = build_zonelists_node(pgdat, zonelist, 0, i); | |
2094 | zonelist->zones[j] = NULL; | |
2095 | } | |
2096 | } | |
2097 | ||
f0c0b2b8 KH |
2098 | /* |
2099 | * Build zonelists ordered by zone and nodes within zones. | |
2100 | * This results in conserving DMA zone[s] until all Normal memory is | |
2101 | * exhausted, but results in overflowing to remote node while memory | |
2102 | * may still exist in local DMA zone. | |
2103 | */ | |
2104 | static int node_order[MAX_NUMNODES]; | |
2105 | ||
2106 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
2107 | { | |
2108 | enum zone_type i; | |
2109 | int pos, j, node; | |
2110 | int zone_type; /* needs to be signed */ | |
2111 | struct zone *z; | |
2112 | struct zonelist *zonelist; | |
2113 | ||
2114 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2115 | zonelist = pgdat->node_zonelists + i; | |
2116 | pos = 0; | |
2117 | for (zone_type = i; zone_type >= 0; zone_type--) { | |
2118 | for (j = 0; j < nr_nodes; j++) { | |
2119 | node = node_order[j]; | |
2120 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
2121 | if (populated_zone(z)) { | |
2122 | zonelist->zones[pos++] = z; | |
2123 | check_highest_zone(zone_type); | |
2124 | } | |
2125 | } | |
2126 | } | |
2127 | zonelist->zones[pos] = NULL; | |
2128 | } | |
2129 | } | |
2130 | ||
2131 | static int default_zonelist_order(void) | |
2132 | { | |
2133 | int nid, zone_type; | |
2134 | unsigned long low_kmem_size,total_size; | |
2135 | struct zone *z; | |
2136 | int average_size; | |
2137 | /* | |
2138 | * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem. | |
2139 | * If they are really small and used heavily, the system can fall | |
2140 | * into OOM very easily. | |
2141 | * This function detect ZONE_DMA/DMA32 size and confgigures zone order. | |
2142 | */ | |
2143 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
2144 | low_kmem_size = 0; | |
2145 | total_size = 0; | |
2146 | for_each_online_node(nid) { | |
2147 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2148 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2149 | if (populated_zone(z)) { | |
2150 | if (zone_type < ZONE_NORMAL) | |
2151 | low_kmem_size += z->present_pages; | |
2152 | total_size += z->present_pages; | |
2153 | } | |
2154 | } | |
2155 | } | |
2156 | if (!low_kmem_size || /* there are no DMA area. */ | |
2157 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
2158 | return ZONELIST_ORDER_NODE; | |
2159 | /* | |
2160 | * look into each node's config. | |
2161 | * If there is a node whose DMA/DMA32 memory is very big area on | |
2162 | * local memory, NODE_ORDER may be suitable. | |
2163 | */ | |
37b07e41 LS |
2164 | average_size = total_size / |
2165 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); | |
f0c0b2b8 KH |
2166 | for_each_online_node(nid) { |
2167 | low_kmem_size = 0; | |
2168 | total_size = 0; | |
2169 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2170 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2171 | if (populated_zone(z)) { | |
2172 | if (zone_type < ZONE_NORMAL) | |
2173 | low_kmem_size += z->present_pages; | |
2174 | total_size += z->present_pages; | |
2175 | } | |
2176 | } | |
2177 | if (low_kmem_size && | |
2178 | total_size > average_size && /* ignore small node */ | |
2179 | low_kmem_size > total_size * 70/100) | |
2180 | return ZONELIST_ORDER_NODE; | |
2181 | } | |
2182 | return ZONELIST_ORDER_ZONE; | |
2183 | } | |
2184 | ||
2185 | static void set_zonelist_order(void) | |
2186 | { | |
2187 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
2188 | current_zonelist_order = default_zonelist_order(); | |
2189 | else | |
2190 | current_zonelist_order = user_zonelist_order; | |
2191 | } | |
2192 | ||
2193 | static void build_zonelists(pg_data_t *pgdat) | |
2194 | { | |
2195 | int j, node, load; | |
2196 | enum zone_type i; | |
1da177e4 | 2197 | nodemask_t used_mask; |
f0c0b2b8 KH |
2198 | int local_node, prev_node; |
2199 | struct zonelist *zonelist; | |
2200 | int order = current_zonelist_order; | |
1da177e4 LT |
2201 | |
2202 | /* initialize zonelists */ | |
523b9458 | 2203 | for (i = 0; i < MAX_ZONELISTS; i++) { |
1da177e4 LT |
2204 | zonelist = pgdat->node_zonelists + i; |
2205 | zonelist->zones[0] = NULL; | |
2206 | } | |
2207 | ||
2208 | /* NUMA-aware ordering of nodes */ | |
2209 | local_node = pgdat->node_id; | |
2210 | load = num_online_nodes(); | |
2211 | prev_node = local_node; | |
2212 | nodes_clear(used_mask); | |
f0c0b2b8 KH |
2213 | |
2214 | memset(node_load, 0, sizeof(node_load)); | |
2215 | memset(node_order, 0, sizeof(node_order)); | |
2216 | j = 0; | |
2217 | ||
1da177e4 | 2218 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
9eeff239 CL |
2219 | int distance = node_distance(local_node, node); |
2220 | ||
2221 | /* | |
2222 | * If another node is sufficiently far away then it is better | |
2223 | * to reclaim pages in a zone before going off node. | |
2224 | */ | |
2225 | if (distance > RECLAIM_DISTANCE) | |
2226 | zone_reclaim_mode = 1; | |
2227 | ||
1da177e4 LT |
2228 | /* |
2229 | * We don't want to pressure a particular node. | |
2230 | * So adding penalty to the first node in same | |
2231 | * distance group to make it round-robin. | |
2232 | */ | |
9eeff239 | 2233 | if (distance != node_distance(local_node, prev_node)) |
f0c0b2b8 KH |
2234 | node_load[node] = load; |
2235 | ||
1da177e4 LT |
2236 | prev_node = node; |
2237 | load--; | |
f0c0b2b8 KH |
2238 | if (order == ZONELIST_ORDER_NODE) |
2239 | build_zonelists_in_node_order(pgdat, node); | |
2240 | else | |
2241 | node_order[j++] = node; /* remember order */ | |
2242 | } | |
1da177e4 | 2243 | |
f0c0b2b8 KH |
2244 | if (order == ZONELIST_ORDER_ZONE) { |
2245 | /* calculate node order -- i.e., DMA last! */ | |
2246 | build_zonelists_in_zone_order(pgdat, j); | |
1da177e4 | 2247 | } |
523b9458 CL |
2248 | |
2249 | build_thisnode_zonelists(pgdat); | |
1da177e4 LT |
2250 | } |
2251 | ||
9276b1bc | 2252 | /* Construct the zonelist performance cache - see further mmzone.h */ |
f0c0b2b8 | 2253 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc PJ |
2254 | { |
2255 | int i; | |
2256 | ||
2257 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
2258 | struct zonelist *zonelist; | |
2259 | struct zonelist_cache *zlc; | |
2260 | struct zone **z; | |
2261 | ||
2262 | zonelist = pgdat->node_zonelists + i; | |
2263 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
2264 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
2265 | for (z = zonelist->zones; *z; z++) | |
2266 | zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z); | |
2267 | } | |
2268 | } | |
2269 | ||
f0c0b2b8 | 2270 | |
1da177e4 LT |
2271 | #else /* CONFIG_NUMA */ |
2272 | ||
f0c0b2b8 KH |
2273 | static void set_zonelist_order(void) |
2274 | { | |
2275 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
2276 | } | |
2277 | ||
2278 | static void build_zonelists(pg_data_t *pgdat) | |
1da177e4 | 2279 | { |
19655d34 CL |
2280 | int node, local_node; |
2281 | enum zone_type i,j; | |
1da177e4 LT |
2282 | |
2283 | local_node = pgdat->node_id; | |
19655d34 | 2284 | for (i = 0; i < MAX_NR_ZONES; i++) { |
1da177e4 LT |
2285 | struct zonelist *zonelist; |
2286 | ||
2287 | zonelist = pgdat->node_zonelists + i; | |
2288 | ||
19655d34 | 2289 | j = build_zonelists_node(pgdat, zonelist, 0, i); |
1da177e4 LT |
2290 | /* |
2291 | * Now we build the zonelist so that it contains the zones | |
2292 | * of all the other nodes. | |
2293 | * We don't want to pressure a particular node, so when | |
2294 | * building the zones for node N, we make sure that the | |
2295 | * zones coming right after the local ones are those from | |
2296 | * node N+1 (modulo N) | |
2297 | */ | |
2298 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
2299 | if (!node_online(node)) | |
2300 | continue; | |
19655d34 | 2301 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
2302 | } |
2303 | for (node = 0; node < local_node; node++) { | |
2304 | if (!node_online(node)) | |
2305 | continue; | |
19655d34 | 2306 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); |
1da177e4 LT |
2307 | } |
2308 | ||
2309 | zonelist->zones[j] = NULL; | |
2310 | } | |
2311 | } | |
2312 | ||
9276b1bc | 2313 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
f0c0b2b8 | 2314 | static void build_zonelist_cache(pg_data_t *pgdat) |
9276b1bc PJ |
2315 | { |
2316 | int i; | |
2317 | ||
2318 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2319 | pgdat->node_zonelists[i].zlcache_ptr = NULL; | |
2320 | } | |
2321 | ||
1da177e4 LT |
2322 | #endif /* CONFIG_NUMA */ |
2323 | ||
6811378e | 2324 | /* return values int ....just for stop_machine_run() */ |
f0c0b2b8 | 2325 | static int __build_all_zonelists(void *dummy) |
1da177e4 | 2326 | { |
6811378e | 2327 | int nid; |
9276b1bc PJ |
2328 | |
2329 | for_each_online_node(nid) { | |
7ea1530a CL |
2330 | pg_data_t *pgdat = NODE_DATA(nid); |
2331 | ||
2332 | build_zonelists(pgdat); | |
2333 | build_zonelist_cache(pgdat); | |
9276b1bc | 2334 | } |
6811378e YG |
2335 | return 0; |
2336 | } | |
2337 | ||
f0c0b2b8 | 2338 | void build_all_zonelists(void) |
6811378e | 2339 | { |
f0c0b2b8 KH |
2340 | set_zonelist_order(); |
2341 | ||
6811378e | 2342 | if (system_state == SYSTEM_BOOTING) { |
423b41d7 | 2343 | __build_all_zonelists(NULL); |
6811378e YG |
2344 | cpuset_init_current_mems_allowed(); |
2345 | } else { | |
183ff22b | 2346 | /* we have to stop all cpus to guarantee there is no user |
6811378e YG |
2347 | of zonelist */ |
2348 | stop_machine_run(__build_all_zonelists, NULL, NR_CPUS); | |
2349 | /* cpuset refresh routine should be here */ | |
2350 | } | |
bd1e22b8 | 2351 | vm_total_pages = nr_free_pagecache_pages(); |
9ef9acb0 MG |
2352 | /* |
2353 | * Disable grouping by mobility if the number of pages in the | |
2354 | * system is too low to allow the mechanism to work. It would be | |
2355 | * more accurate, but expensive to check per-zone. This check is | |
2356 | * made on memory-hotadd so a system can start with mobility | |
2357 | * disabled and enable it later | |
2358 | */ | |
d9c23400 | 2359 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
9ef9acb0 MG |
2360 | page_group_by_mobility_disabled = 1; |
2361 | else | |
2362 | page_group_by_mobility_disabled = 0; | |
2363 | ||
2364 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
2365 | "Total pages: %ld\n", | |
f0c0b2b8 KH |
2366 | num_online_nodes(), |
2367 | zonelist_order_name[current_zonelist_order], | |
9ef9acb0 | 2368 | page_group_by_mobility_disabled ? "off" : "on", |
f0c0b2b8 KH |
2369 | vm_total_pages); |
2370 | #ifdef CONFIG_NUMA | |
2371 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
2372 | #endif | |
1da177e4 LT |
2373 | } |
2374 | ||
2375 | /* | |
2376 | * Helper functions to size the waitqueue hash table. | |
2377 | * Essentially these want to choose hash table sizes sufficiently | |
2378 | * large so that collisions trying to wait on pages are rare. | |
2379 | * But in fact, the number of active page waitqueues on typical | |
2380 | * systems is ridiculously low, less than 200. So this is even | |
2381 | * conservative, even though it seems large. | |
2382 | * | |
2383 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
2384 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
2385 | */ | |
2386 | #define PAGES_PER_WAITQUEUE 256 | |
2387 | ||
cca448fe | 2388 | #ifndef CONFIG_MEMORY_HOTPLUG |
02b694de | 2389 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
1da177e4 LT |
2390 | { |
2391 | unsigned long size = 1; | |
2392 | ||
2393 | pages /= PAGES_PER_WAITQUEUE; | |
2394 | ||
2395 | while (size < pages) | |
2396 | size <<= 1; | |
2397 | ||
2398 | /* | |
2399 | * Once we have dozens or even hundreds of threads sleeping | |
2400 | * on IO we've got bigger problems than wait queue collision. | |
2401 | * Limit the size of the wait table to a reasonable size. | |
2402 | */ | |
2403 | size = min(size, 4096UL); | |
2404 | ||
2405 | return max(size, 4UL); | |
2406 | } | |
cca448fe YG |
2407 | #else |
2408 | /* | |
2409 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
2410 | * a suitable size for its wait_table. So we use the maximum size now. | |
2411 | * | |
2412 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
2413 | * | |
2414 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
2415 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
2416 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
2417 | * | |
2418 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
2419 | * or more by the traditional way. (See above). It equals: | |
2420 | * | |
2421 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
2422 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
2423 | * powerpc (64K page size) : = (32G +16M)byte. | |
2424 | */ | |
2425 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
2426 | { | |
2427 | return 4096UL; | |
2428 | } | |
2429 | #endif | |
1da177e4 LT |
2430 | |
2431 | /* | |
2432 | * This is an integer logarithm so that shifts can be used later | |
2433 | * to extract the more random high bits from the multiplicative | |
2434 | * hash function before the remainder is taken. | |
2435 | */ | |
2436 | static inline unsigned long wait_table_bits(unsigned long size) | |
2437 | { | |
2438 | return ffz(~size); | |
2439 | } | |
2440 | ||
2441 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
2442 | ||
56fd56b8 | 2443 | /* |
d9c23400 | 2444 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
56fd56b8 MG |
2445 | * of blocks reserved is based on zone->pages_min. The memory within the |
2446 | * reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
2447 | * higher will lead to a bigger reserve which will get freed as contiguous | |
2448 | * blocks as reclaim kicks in | |
2449 | */ | |
2450 | static void setup_zone_migrate_reserve(struct zone *zone) | |
2451 | { | |
2452 | unsigned long start_pfn, pfn, end_pfn; | |
2453 | struct page *page; | |
2454 | unsigned long reserve, block_migratetype; | |
2455 | ||
2456 | /* Get the start pfn, end pfn and the number of blocks to reserve */ | |
2457 | start_pfn = zone->zone_start_pfn; | |
2458 | end_pfn = start_pfn + zone->spanned_pages; | |
d9c23400 MG |
2459 | reserve = roundup(zone->pages_min, pageblock_nr_pages) >> |
2460 | pageblock_order; | |
56fd56b8 | 2461 | |
d9c23400 | 2462 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
56fd56b8 MG |
2463 | if (!pfn_valid(pfn)) |
2464 | continue; | |
2465 | page = pfn_to_page(pfn); | |
2466 | ||
2467 | /* Blocks with reserved pages will never free, skip them. */ | |
2468 | if (PageReserved(page)) | |
2469 | continue; | |
2470 | ||
2471 | block_migratetype = get_pageblock_migratetype(page); | |
2472 | ||
2473 | /* If this block is reserved, account for it */ | |
2474 | if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) { | |
2475 | reserve--; | |
2476 | continue; | |
2477 | } | |
2478 | ||
2479 | /* Suitable for reserving if this block is movable */ | |
2480 | if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) { | |
2481 | set_pageblock_migratetype(page, MIGRATE_RESERVE); | |
2482 | move_freepages_block(zone, page, MIGRATE_RESERVE); | |
2483 | reserve--; | |
2484 | continue; | |
2485 | } | |
2486 | ||
2487 | /* | |
2488 | * If the reserve is met and this is a previous reserved block, | |
2489 | * take it back | |
2490 | */ | |
2491 | if (block_migratetype == MIGRATE_RESERVE) { | |
2492 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
2493 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
2494 | } | |
2495 | } | |
2496 | } | |
ac0e5b7a | 2497 | |
1da177e4 LT |
2498 | /* |
2499 | * Initially all pages are reserved - free ones are freed | |
2500 | * up by free_all_bootmem() once the early boot process is | |
2501 | * done. Non-atomic initialization, single-pass. | |
2502 | */ | |
c09b4240 | 2503 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
a2f3aa02 | 2504 | unsigned long start_pfn, enum memmap_context context) |
1da177e4 | 2505 | { |
1da177e4 | 2506 | struct page *page; |
29751f69 AW |
2507 | unsigned long end_pfn = start_pfn + size; |
2508 | unsigned long pfn; | |
1da177e4 | 2509 | |
cbe8dd4a | 2510 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
a2f3aa02 DH |
2511 | /* |
2512 | * There can be holes in boot-time mem_map[]s | |
2513 | * handed to this function. They do not | |
2514 | * exist on hotplugged memory. | |
2515 | */ | |
2516 | if (context == MEMMAP_EARLY) { | |
2517 | if (!early_pfn_valid(pfn)) | |
2518 | continue; | |
2519 | if (!early_pfn_in_nid(pfn, nid)) | |
2520 | continue; | |
2521 | } | |
d41dee36 AW |
2522 | page = pfn_to_page(pfn); |
2523 | set_page_links(page, zone, nid, pfn); | |
7835e98b | 2524 | init_page_count(page); |
1da177e4 LT |
2525 | reset_page_mapcount(page); |
2526 | SetPageReserved(page); | |
b2a0ac88 MG |
2527 | |
2528 | /* | |
2529 | * Mark the block movable so that blocks are reserved for | |
2530 | * movable at startup. This will force kernel allocations | |
2531 | * to reserve their blocks rather than leaking throughout | |
2532 | * the address space during boot when many long-lived | |
56fd56b8 MG |
2533 | * kernel allocations are made. Later some blocks near |
2534 | * the start are marked MIGRATE_RESERVE by | |
2535 | * setup_zone_migrate_reserve() | |
b2a0ac88 | 2536 | */ |
d9c23400 | 2537 | if ((pfn & (pageblock_nr_pages-1))) |
56fd56b8 | 2538 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
b2a0ac88 | 2539 | |
1da177e4 LT |
2540 | INIT_LIST_HEAD(&page->lru); |
2541 | #ifdef WANT_PAGE_VIRTUAL | |
2542 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
2543 | if (!is_highmem_idx(zone)) | |
3212c6be | 2544 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 2545 | #endif |
1da177e4 LT |
2546 | } |
2547 | } | |
2548 | ||
1e548deb | 2549 | static void __meminit zone_init_free_lists(struct zone *zone) |
1da177e4 | 2550 | { |
b2a0ac88 MG |
2551 | int order, t; |
2552 | for_each_migratetype_order(order, t) { | |
2553 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
1da177e4 LT |
2554 | zone->free_area[order].nr_free = 0; |
2555 | } | |
2556 | } | |
2557 | ||
2558 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
2559 | #define memmap_init(size, nid, zone, start_pfn) \ | |
a2f3aa02 | 2560 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
1da177e4 LT |
2561 | #endif |
2562 | ||
1d6f4e60 | 2563 | static int zone_batchsize(struct zone *zone) |
e7c8d5c9 CL |
2564 | { |
2565 | int batch; | |
2566 | ||
2567 | /* | |
2568 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 2569 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
2570 | * |
2571 | * OK, so we don't know how big the cache is. So guess. | |
2572 | */ | |
2573 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
2574 | if (batch * PAGE_SIZE > 512 * 1024) |
2575 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
2576 | batch /= 4; /* We effectively *= 4 below */ |
2577 | if (batch < 1) | |
2578 | batch = 1; | |
2579 | ||
2580 | /* | |
0ceaacc9 NP |
2581 | * Clamp the batch to a 2^n - 1 value. Having a power |
2582 | * of 2 value was found to be more likely to have | |
2583 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 2584 | * |
0ceaacc9 NP |
2585 | * For example if 2 tasks are alternately allocating |
2586 | * batches of pages, one task can end up with a lot | |
2587 | * of pages of one half of the possible page colors | |
2588 | * and the other with pages of the other colors. | |
e7c8d5c9 | 2589 | */ |
0ceaacc9 | 2590 | batch = (1 << (fls(batch + batch/2)-1)) - 1; |
ba56e91c | 2591 | |
e7c8d5c9 CL |
2592 | return batch; |
2593 | } | |
2594 | ||
2caaad41 CL |
2595 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
2596 | { | |
2597 | struct per_cpu_pages *pcp; | |
2598 | ||
1c6fe946 MD |
2599 | memset(p, 0, sizeof(*p)); |
2600 | ||
3dfa5721 | 2601 | pcp = &p->pcp; |
2caaad41 | 2602 | pcp->count = 0; |
2caaad41 CL |
2603 | pcp->high = 6 * batch; |
2604 | pcp->batch = max(1UL, 1 * batch); | |
2605 | INIT_LIST_HEAD(&pcp->list); | |
2caaad41 CL |
2606 | } |
2607 | ||
8ad4b1fb RS |
2608 | /* |
2609 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
2610 | * to the value high for the pageset p. | |
2611 | */ | |
2612 | ||
2613 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
2614 | unsigned long high) | |
2615 | { | |
2616 | struct per_cpu_pages *pcp; | |
2617 | ||
3dfa5721 | 2618 | pcp = &p->pcp; |
8ad4b1fb RS |
2619 | pcp->high = high; |
2620 | pcp->batch = max(1UL, high/4); | |
2621 | if ((high/4) > (PAGE_SHIFT * 8)) | |
2622 | pcp->batch = PAGE_SHIFT * 8; | |
2623 | } | |
2624 | ||
2625 | ||
e7c8d5c9 CL |
2626 | #ifdef CONFIG_NUMA |
2627 | /* | |
2caaad41 CL |
2628 | * Boot pageset table. One per cpu which is going to be used for all |
2629 | * zones and all nodes. The parameters will be set in such a way | |
2630 | * that an item put on a list will immediately be handed over to | |
2631 | * the buddy list. This is safe since pageset manipulation is done | |
2632 | * with interrupts disabled. | |
2633 | * | |
2634 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
2635 | * |
2636 | * The boot_pagesets must be kept even after bootup is complete for | |
2637 | * unused processors and/or zones. They do play a role for bootstrapping | |
2638 | * hotplugged processors. | |
2639 | * | |
2640 | * zoneinfo_show() and maybe other functions do | |
2641 | * not check if the processor is online before following the pageset pointer. | |
2642 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 2643 | */ |
88a2a4ac | 2644 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
2645 | |
2646 | /* | |
2647 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
2648 | * per cpu pageset array in struct zone. |
2649 | */ | |
6292d9aa | 2650 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
2651 | { |
2652 | struct zone *zone, *dzone; | |
37c0708d CL |
2653 | int node = cpu_to_node(cpu); |
2654 | ||
2655 | node_set_state(node, N_CPU); /* this node has a cpu */ | |
e7c8d5c9 CL |
2656 | |
2657 | for_each_zone(zone) { | |
e7c8d5c9 | 2658 | |
66a55030 CL |
2659 | if (!populated_zone(zone)) |
2660 | continue; | |
2661 | ||
23316bc8 | 2662 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
37c0708d | 2663 | GFP_KERNEL, node); |
23316bc8 | 2664 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 2665 | goto bad; |
e7c8d5c9 | 2666 | |
23316bc8 | 2667 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
2668 | |
2669 | if (percpu_pagelist_fraction) | |
2670 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
2671 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
2672 | } |
2673 | ||
2674 | return 0; | |
2675 | bad: | |
2676 | for_each_zone(dzone) { | |
64191688 AM |
2677 | if (!populated_zone(dzone)) |
2678 | continue; | |
e7c8d5c9 CL |
2679 | if (dzone == zone) |
2680 | break; | |
23316bc8 NP |
2681 | kfree(zone_pcp(dzone, cpu)); |
2682 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
2683 | } |
2684 | return -ENOMEM; | |
2685 | } | |
2686 | ||
2687 | static inline void free_zone_pagesets(int cpu) | |
2688 | { | |
e7c8d5c9 CL |
2689 | struct zone *zone; |
2690 | ||
2691 | for_each_zone(zone) { | |
2692 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
2693 | ||
f3ef9ead DR |
2694 | /* Free per_cpu_pageset if it is slab allocated */ |
2695 | if (pset != &boot_pageset[cpu]) | |
2696 | kfree(pset); | |
e7c8d5c9 | 2697 | zone_pcp(zone, cpu) = NULL; |
e7c8d5c9 | 2698 | } |
e7c8d5c9 CL |
2699 | } |
2700 | ||
9c7b216d | 2701 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
2702 | unsigned long action, |
2703 | void *hcpu) | |
2704 | { | |
2705 | int cpu = (long)hcpu; | |
2706 | int ret = NOTIFY_OK; | |
2707 | ||
2708 | switch (action) { | |
ce421c79 | 2709 | case CPU_UP_PREPARE: |
8bb78442 | 2710 | case CPU_UP_PREPARE_FROZEN: |
ce421c79 AW |
2711 | if (process_zones(cpu)) |
2712 | ret = NOTIFY_BAD; | |
2713 | break; | |
2714 | case CPU_UP_CANCELED: | |
8bb78442 | 2715 | case CPU_UP_CANCELED_FROZEN: |
ce421c79 | 2716 | case CPU_DEAD: |
8bb78442 | 2717 | case CPU_DEAD_FROZEN: |
ce421c79 AW |
2718 | free_zone_pagesets(cpu); |
2719 | break; | |
2720 | default: | |
2721 | break; | |
e7c8d5c9 CL |
2722 | } |
2723 | return ret; | |
2724 | } | |
2725 | ||
74b85f37 | 2726 | static struct notifier_block __cpuinitdata pageset_notifier = |
e7c8d5c9 CL |
2727 | { &pageset_cpuup_callback, NULL, 0 }; |
2728 | ||
78d9955b | 2729 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
2730 | { |
2731 | int err; | |
2732 | ||
2733 | /* Initialize per_cpu_pageset for cpu 0. | |
2734 | * A cpuup callback will do this for every cpu | |
2735 | * as it comes online | |
2736 | */ | |
2737 | err = process_zones(smp_processor_id()); | |
2738 | BUG_ON(err); | |
2739 | register_cpu_notifier(&pageset_notifier); | |
2740 | } | |
2741 | ||
2742 | #endif | |
2743 | ||
577a32f6 | 2744 | static noinline __init_refok |
cca448fe | 2745 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
ed8ece2e DH |
2746 | { |
2747 | int i; | |
2748 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe | 2749 | size_t alloc_size; |
ed8ece2e DH |
2750 | |
2751 | /* | |
2752 | * The per-page waitqueue mechanism uses hashed waitqueues | |
2753 | * per zone. | |
2754 | */ | |
02b694de YG |
2755 | zone->wait_table_hash_nr_entries = |
2756 | wait_table_hash_nr_entries(zone_size_pages); | |
2757 | zone->wait_table_bits = | |
2758 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
cca448fe YG |
2759 | alloc_size = zone->wait_table_hash_nr_entries |
2760 | * sizeof(wait_queue_head_t); | |
2761 | ||
2762 | if (system_state == SYSTEM_BOOTING) { | |
2763 | zone->wait_table = (wait_queue_head_t *) | |
2764 | alloc_bootmem_node(pgdat, alloc_size); | |
2765 | } else { | |
2766 | /* | |
2767 | * This case means that a zone whose size was 0 gets new memory | |
2768 | * via memory hot-add. | |
2769 | * But it may be the case that a new node was hot-added. In | |
2770 | * this case vmalloc() will not be able to use this new node's | |
2771 | * memory - this wait_table must be initialized to use this new | |
2772 | * node itself as well. | |
2773 | * To use this new node's memory, further consideration will be | |
2774 | * necessary. | |
2775 | */ | |
8691f3a7 | 2776 | zone->wait_table = vmalloc(alloc_size); |
cca448fe YG |
2777 | } |
2778 | if (!zone->wait_table) | |
2779 | return -ENOMEM; | |
ed8ece2e | 2780 | |
02b694de | 2781 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
ed8ece2e | 2782 | init_waitqueue_head(zone->wait_table + i); |
cca448fe YG |
2783 | |
2784 | return 0; | |
ed8ece2e DH |
2785 | } |
2786 | ||
c09b4240 | 2787 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
2788 | { |
2789 | int cpu; | |
2790 | unsigned long batch = zone_batchsize(zone); | |
2791 | ||
2792 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
2793 | #ifdef CONFIG_NUMA | |
2794 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 2795 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
2796 | setup_pageset(&boot_pageset[cpu],0); |
2797 | #else | |
2798 | setup_pageset(zone_pcp(zone,cpu), batch); | |
2799 | #endif | |
2800 | } | |
f5335c0f AB |
2801 | if (zone->present_pages) |
2802 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
2803 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
2804 | } |
2805 | ||
718127cc YG |
2806 | __meminit int init_currently_empty_zone(struct zone *zone, |
2807 | unsigned long zone_start_pfn, | |
a2f3aa02 DH |
2808 | unsigned long size, |
2809 | enum memmap_context context) | |
ed8ece2e DH |
2810 | { |
2811 | struct pglist_data *pgdat = zone->zone_pgdat; | |
cca448fe YG |
2812 | int ret; |
2813 | ret = zone_wait_table_init(zone, size); | |
2814 | if (ret) | |
2815 | return ret; | |
ed8ece2e DH |
2816 | pgdat->nr_zones = zone_idx(zone) + 1; |
2817 | ||
ed8ece2e DH |
2818 | zone->zone_start_pfn = zone_start_pfn; |
2819 | ||
2820 | memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn); | |
2821 | ||
1e548deb | 2822 | zone_init_free_lists(zone); |
718127cc YG |
2823 | |
2824 | return 0; | |
ed8ece2e DH |
2825 | } |
2826 | ||
c713216d MG |
2827 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
2828 | /* | |
2829 | * Basic iterator support. Return the first range of PFNs for a node | |
2830 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
2831 | */ | |
a3142c8e | 2832 | static int __meminit first_active_region_index_in_nid(int nid) |
c713216d MG |
2833 | { |
2834 | int i; | |
2835 | ||
2836 | for (i = 0; i < nr_nodemap_entries; i++) | |
2837 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
2838 | return i; | |
2839 | ||
2840 | return -1; | |
2841 | } | |
2842 | ||
2843 | /* | |
2844 | * Basic iterator support. Return the next active range of PFNs for a node | |
183ff22b | 2845 | * Note: nid == MAX_NUMNODES returns next region regardless of node |
c713216d | 2846 | */ |
a3142c8e | 2847 | static int __meminit next_active_region_index_in_nid(int index, int nid) |
c713216d MG |
2848 | { |
2849 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
2850 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
2851 | return index; | |
2852 | ||
2853 | return -1; | |
2854 | } | |
2855 | ||
2856 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
2857 | /* | |
2858 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
2859 | * Architectures may implement their own version but if add_active_range() | |
2860 | * was used and there are no special requirements, this is a convenient | |
2861 | * alternative | |
2862 | */ | |
6f076f5d | 2863 | int __meminit early_pfn_to_nid(unsigned long pfn) |
c713216d MG |
2864 | { |
2865 | int i; | |
2866 | ||
2867 | for (i = 0; i < nr_nodemap_entries; i++) { | |
2868 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
2869 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2870 | ||
2871 | if (start_pfn <= pfn && pfn < end_pfn) | |
2872 | return early_node_map[i].nid; | |
2873 | } | |
2874 | ||
2875 | return 0; | |
2876 | } | |
2877 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
2878 | ||
2879 | /* Basic iterator support to walk early_node_map[] */ | |
2880 | #define for_each_active_range_index_in_nid(i, nid) \ | |
2881 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
2882 | i = next_active_region_index_in_nid(i, nid)) | |
2883 | ||
2884 | /** | |
2885 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
88ca3b94 RD |
2886 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
2887 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
c713216d MG |
2888 | * |
2889 | * If an architecture guarantees that all ranges registered with | |
2890 | * add_active_ranges() contain no holes and may be freed, this | |
2891 | * this function may be used instead of calling free_bootmem() manually. | |
2892 | */ | |
2893 | void __init free_bootmem_with_active_regions(int nid, | |
2894 | unsigned long max_low_pfn) | |
2895 | { | |
2896 | int i; | |
2897 | ||
2898 | for_each_active_range_index_in_nid(i, nid) { | |
2899 | unsigned long size_pages = 0; | |
2900 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2901 | ||
2902 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
2903 | continue; | |
2904 | ||
2905 | if (end_pfn > max_low_pfn) | |
2906 | end_pfn = max_low_pfn; | |
2907 | ||
2908 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
2909 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
2910 | PFN_PHYS(early_node_map[i].start_pfn), | |
2911 | size_pages << PAGE_SHIFT); | |
2912 | } | |
2913 | } | |
2914 | ||
2915 | /** | |
2916 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
88ca3b94 | 2917 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
c713216d MG |
2918 | * |
2919 | * If an architecture guarantees that all ranges registered with | |
2920 | * add_active_ranges() contain no holes and may be freed, this | |
88ca3b94 | 2921 | * function may be used instead of calling memory_present() manually. |
c713216d MG |
2922 | */ |
2923 | void __init sparse_memory_present_with_active_regions(int nid) | |
2924 | { | |
2925 | int i; | |
2926 | ||
2927 | for_each_active_range_index_in_nid(i, nid) | |
2928 | memory_present(early_node_map[i].nid, | |
2929 | early_node_map[i].start_pfn, | |
2930 | early_node_map[i].end_pfn); | |
2931 | } | |
2932 | ||
fb01439c MG |
2933 | /** |
2934 | * push_node_boundaries - Push node boundaries to at least the requested boundary | |
2935 | * @nid: The nid of the node to push the boundary for | |
2936 | * @start_pfn: The start pfn of the node | |
2937 | * @end_pfn: The end pfn of the node | |
2938 | * | |
2939 | * In reserve-based hot-add, mem_map is allocated that is unused until hotadd | |
2940 | * time. Specifically, on x86_64, SRAT will report ranges that can potentially | |
2941 | * be hotplugged even though no physical memory exists. This function allows | |
2942 | * an arch to push out the node boundaries so mem_map is allocated that can | |
2943 | * be used later. | |
2944 | */ | |
2945 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE | |
2946 | void __init push_node_boundaries(unsigned int nid, | |
2947 | unsigned long start_pfn, unsigned long end_pfn) | |
2948 | { | |
2949 | printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n", | |
2950 | nid, start_pfn, end_pfn); | |
2951 | ||
2952 | /* Initialise the boundary for this node if necessary */ | |
2953 | if (node_boundary_end_pfn[nid] == 0) | |
2954 | node_boundary_start_pfn[nid] = -1UL; | |
2955 | ||
2956 | /* Update the boundaries */ | |
2957 | if (node_boundary_start_pfn[nid] > start_pfn) | |
2958 | node_boundary_start_pfn[nid] = start_pfn; | |
2959 | if (node_boundary_end_pfn[nid] < end_pfn) | |
2960 | node_boundary_end_pfn[nid] = end_pfn; | |
2961 | } | |
2962 | ||
2963 | /* If necessary, push the node boundary out for reserve hotadd */ | |
98011f56 | 2964 | static void __meminit account_node_boundary(unsigned int nid, |
fb01439c MG |
2965 | unsigned long *start_pfn, unsigned long *end_pfn) |
2966 | { | |
2967 | printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n", | |
2968 | nid, *start_pfn, *end_pfn); | |
2969 | ||
2970 | /* Return if boundary information has not been provided */ | |
2971 | if (node_boundary_end_pfn[nid] == 0) | |
2972 | return; | |
2973 | ||
2974 | /* Check the boundaries and update if necessary */ | |
2975 | if (node_boundary_start_pfn[nid] < *start_pfn) | |
2976 | *start_pfn = node_boundary_start_pfn[nid]; | |
2977 | if (node_boundary_end_pfn[nid] > *end_pfn) | |
2978 | *end_pfn = node_boundary_end_pfn[nid]; | |
2979 | } | |
2980 | #else | |
2981 | void __init push_node_boundaries(unsigned int nid, | |
2982 | unsigned long start_pfn, unsigned long end_pfn) {} | |
2983 | ||
98011f56 | 2984 | static void __meminit account_node_boundary(unsigned int nid, |
fb01439c MG |
2985 | unsigned long *start_pfn, unsigned long *end_pfn) {} |
2986 | #endif | |
2987 | ||
2988 | ||
c713216d MG |
2989 | /** |
2990 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
88ca3b94 RD |
2991 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
2992 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
2993 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
c713216d MG |
2994 | * |
2995 | * It returns the start and end page frame of a node based on information | |
2996 | * provided by an arch calling add_active_range(). If called for a node | |
2997 | * with no available memory, a warning is printed and the start and end | |
88ca3b94 | 2998 | * PFNs will be 0. |
c713216d | 2999 | */ |
a3142c8e | 3000 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
c713216d MG |
3001 | unsigned long *start_pfn, unsigned long *end_pfn) |
3002 | { | |
3003 | int i; | |
3004 | *start_pfn = -1UL; | |
3005 | *end_pfn = 0; | |
3006 | ||
3007 | for_each_active_range_index_in_nid(i, nid) { | |
3008 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
3009 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
3010 | } | |
3011 | ||
633c0666 | 3012 | if (*start_pfn == -1UL) |
c713216d | 3013 | *start_pfn = 0; |
fb01439c MG |
3014 | |
3015 | /* Push the node boundaries out if requested */ | |
3016 | account_node_boundary(nid, start_pfn, end_pfn); | |
c713216d MG |
3017 | } |
3018 | ||
2a1e274a MG |
3019 | /* |
3020 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
3021 | * assumption is made that zones within a node are ordered in monotonic | |
3022 | * increasing memory addresses so that the "highest" populated zone is used | |
3023 | */ | |
3024 | void __init find_usable_zone_for_movable(void) | |
3025 | { | |
3026 | int zone_index; | |
3027 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
3028 | if (zone_index == ZONE_MOVABLE) | |
3029 | continue; | |
3030 | ||
3031 | if (arch_zone_highest_possible_pfn[zone_index] > | |
3032 | arch_zone_lowest_possible_pfn[zone_index]) | |
3033 | break; | |
3034 | } | |
3035 | ||
3036 | VM_BUG_ON(zone_index == -1); | |
3037 | movable_zone = zone_index; | |
3038 | } | |
3039 | ||
3040 | /* | |
3041 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
3042 | * because it is sized independant of architecture. Unlike the other zones, | |
3043 | * the starting point for ZONE_MOVABLE is not fixed. It may be different | |
3044 | * in each node depending on the size of each node and how evenly kernelcore | |
3045 | * is distributed. This helper function adjusts the zone ranges | |
3046 | * provided by the architecture for a given node by using the end of the | |
3047 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
3048 | * zones within a node are in order of monotonic increases memory addresses | |
3049 | */ | |
3050 | void __meminit adjust_zone_range_for_zone_movable(int nid, | |
3051 | unsigned long zone_type, | |
3052 | unsigned long node_start_pfn, | |
3053 | unsigned long node_end_pfn, | |
3054 | unsigned long *zone_start_pfn, | |
3055 | unsigned long *zone_end_pfn) | |
3056 | { | |
3057 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
3058 | if (zone_movable_pfn[nid]) { | |
3059 | /* Size ZONE_MOVABLE */ | |
3060 | if (zone_type == ZONE_MOVABLE) { | |
3061 | *zone_start_pfn = zone_movable_pfn[nid]; | |
3062 | *zone_end_pfn = min(node_end_pfn, | |
3063 | arch_zone_highest_possible_pfn[movable_zone]); | |
3064 | ||
3065 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
3066 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
3067 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
3068 | *zone_end_pfn = zone_movable_pfn[nid]; | |
3069 | ||
3070 | /* Check if this whole range is within ZONE_MOVABLE */ | |
3071 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
3072 | *zone_start_pfn = *zone_end_pfn; | |
3073 | } | |
3074 | } | |
3075 | ||
c713216d MG |
3076 | /* |
3077 | * Return the number of pages a zone spans in a node, including holes | |
3078 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
3079 | */ | |
6ea6e688 | 3080 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3081 | unsigned long zone_type, |
3082 | unsigned long *ignored) | |
3083 | { | |
3084 | unsigned long node_start_pfn, node_end_pfn; | |
3085 | unsigned long zone_start_pfn, zone_end_pfn; | |
3086 | ||
3087 | /* Get the start and end of the node and zone */ | |
3088 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3089 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
3090 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2a1e274a MG |
3091 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3092 | node_start_pfn, node_end_pfn, | |
3093 | &zone_start_pfn, &zone_end_pfn); | |
c713216d MG |
3094 | |
3095 | /* Check that this node has pages within the zone's required range */ | |
3096 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
3097 | return 0; | |
3098 | ||
3099 | /* Move the zone boundaries inside the node if necessary */ | |
3100 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
3101 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
3102 | ||
3103 | /* Return the spanned pages */ | |
3104 | return zone_end_pfn - zone_start_pfn; | |
3105 | } | |
3106 | ||
3107 | /* | |
3108 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
88ca3b94 | 3109 | * then all holes in the requested range will be accounted for. |
c713216d | 3110 | */ |
a3142c8e | 3111 | unsigned long __meminit __absent_pages_in_range(int nid, |
c713216d MG |
3112 | unsigned long range_start_pfn, |
3113 | unsigned long range_end_pfn) | |
3114 | { | |
3115 | int i = 0; | |
3116 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
3117 | unsigned long start_pfn; | |
3118 | ||
3119 | /* Find the end_pfn of the first active range of pfns in the node */ | |
3120 | i = first_active_region_index_in_nid(nid); | |
3121 | if (i == -1) | |
3122 | return 0; | |
3123 | ||
b5445f95 MG |
3124 | prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn); |
3125 | ||
9c7cd687 MG |
3126 | /* Account for ranges before physical memory on this node */ |
3127 | if (early_node_map[i].start_pfn > range_start_pfn) | |
b5445f95 | 3128 | hole_pages = prev_end_pfn - range_start_pfn; |
c713216d MG |
3129 | |
3130 | /* Find all holes for the zone within the node */ | |
3131 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
3132 | ||
3133 | /* No need to continue if prev_end_pfn is outside the zone */ | |
3134 | if (prev_end_pfn >= range_end_pfn) | |
3135 | break; | |
3136 | ||
3137 | /* Make sure the end of the zone is not within the hole */ | |
3138 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
3139 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
3140 | ||
3141 | /* Update the hole size cound and move on */ | |
3142 | if (start_pfn > range_start_pfn) { | |
3143 | BUG_ON(prev_end_pfn > start_pfn); | |
3144 | hole_pages += start_pfn - prev_end_pfn; | |
3145 | } | |
3146 | prev_end_pfn = early_node_map[i].end_pfn; | |
3147 | } | |
3148 | ||
9c7cd687 MG |
3149 | /* Account for ranges past physical memory on this node */ |
3150 | if (range_end_pfn > prev_end_pfn) | |
0c6cb974 | 3151 | hole_pages += range_end_pfn - |
9c7cd687 MG |
3152 | max(range_start_pfn, prev_end_pfn); |
3153 | ||
c713216d MG |
3154 | return hole_pages; |
3155 | } | |
3156 | ||
3157 | /** | |
3158 | * absent_pages_in_range - Return number of page frames in holes within a range | |
3159 | * @start_pfn: The start PFN to start searching for holes | |
3160 | * @end_pfn: The end PFN to stop searching for holes | |
3161 | * | |
88ca3b94 | 3162 | * It returns the number of pages frames in memory holes within a range. |
c713216d MG |
3163 | */ |
3164 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
3165 | unsigned long end_pfn) | |
3166 | { | |
3167 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
3168 | } | |
3169 | ||
3170 | /* Return the number of page frames in holes in a zone on a node */ | |
6ea6e688 | 3171 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3172 | unsigned long zone_type, |
3173 | unsigned long *ignored) | |
3174 | { | |
9c7cd687 MG |
3175 | unsigned long node_start_pfn, node_end_pfn; |
3176 | unsigned long zone_start_pfn, zone_end_pfn; | |
3177 | ||
3178 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3179 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
3180 | node_start_pfn); | |
3181 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
3182 | node_end_pfn); | |
3183 | ||
2a1e274a MG |
3184 | adjust_zone_range_for_zone_movable(nid, zone_type, |
3185 | node_start_pfn, node_end_pfn, | |
3186 | &zone_start_pfn, &zone_end_pfn); | |
9c7cd687 | 3187 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
c713216d | 3188 | } |
0e0b864e | 3189 | |
c713216d | 3190 | #else |
6ea6e688 | 3191 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
c713216d MG |
3192 | unsigned long zone_type, |
3193 | unsigned long *zones_size) | |
3194 | { | |
3195 | return zones_size[zone_type]; | |
3196 | } | |
3197 | ||
6ea6e688 | 3198 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
c713216d MG |
3199 | unsigned long zone_type, |
3200 | unsigned long *zholes_size) | |
3201 | { | |
3202 | if (!zholes_size) | |
3203 | return 0; | |
3204 | ||
3205 | return zholes_size[zone_type]; | |
3206 | } | |
0e0b864e | 3207 | |
c713216d MG |
3208 | #endif |
3209 | ||
a3142c8e | 3210 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
c713216d MG |
3211 | unsigned long *zones_size, unsigned long *zholes_size) |
3212 | { | |
3213 | unsigned long realtotalpages, totalpages = 0; | |
3214 | enum zone_type i; | |
3215 | ||
3216 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3217 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
3218 | zones_size); | |
3219 | pgdat->node_spanned_pages = totalpages; | |
3220 | ||
3221 | realtotalpages = totalpages; | |
3222 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3223 | realtotalpages -= | |
3224 | zone_absent_pages_in_node(pgdat->node_id, i, | |
3225 | zholes_size); | |
3226 | pgdat->node_present_pages = realtotalpages; | |
3227 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
3228 | realtotalpages); | |
3229 | } | |
3230 | ||
835c134e MG |
3231 | #ifndef CONFIG_SPARSEMEM |
3232 | /* | |
3233 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
d9c23400 MG |
3234 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
3235 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
835c134e MG |
3236 | * round what is now in bits to nearest long in bits, then return it in |
3237 | * bytes. | |
3238 | */ | |
3239 | static unsigned long __init usemap_size(unsigned long zonesize) | |
3240 | { | |
3241 | unsigned long usemapsize; | |
3242 | ||
d9c23400 MG |
3243 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
3244 | usemapsize = usemapsize >> pageblock_order; | |
835c134e MG |
3245 | usemapsize *= NR_PAGEBLOCK_BITS; |
3246 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
3247 | ||
3248 | return usemapsize / 8; | |
3249 | } | |
3250 | ||
3251 | static void __init setup_usemap(struct pglist_data *pgdat, | |
3252 | struct zone *zone, unsigned long zonesize) | |
3253 | { | |
3254 | unsigned long usemapsize = usemap_size(zonesize); | |
3255 | zone->pageblock_flags = NULL; | |
3256 | if (usemapsize) { | |
3257 | zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize); | |
3258 | memset(zone->pageblock_flags, 0, usemapsize); | |
3259 | } | |
3260 | } | |
3261 | #else | |
3262 | static void inline setup_usemap(struct pglist_data *pgdat, | |
3263 | struct zone *zone, unsigned long zonesize) {} | |
3264 | #endif /* CONFIG_SPARSEMEM */ | |
3265 | ||
d9c23400 | 3266 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
ba72cb8c MG |
3267 | |
3268 | /* Return a sensible default order for the pageblock size. */ | |
3269 | static inline int pageblock_default_order(void) | |
3270 | { | |
3271 | if (HPAGE_SHIFT > PAGE_SHIFT) | |
3272 | return HUGETLB_PAGE_ORDER; | |
3273 | ||
3274 | return MAX_ORDER-1; | |
3275 | } | |
3276 | ||
d9c23400 MG |
3277 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
3278 | static inline void __init set_pageblock_order(unsigned int order) | |
3279 | { | |
3280 | /* Check that pageblock_nr_pages has not already been setup */ | |
3281 | if (pageblock_order) | |
3282 | return; | |
3283 | ||
3284 | /* | |
3285 | * Assume the largest contiguous order of interest is a huge page. | |
3286 | * This value may be variable depending on boot parameters on IA64 | |
3287 | */ | |
3288 | pageblock_order = order; | |
3289 | } | |
3290 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
3291 | ||
ba72cb8c MG |
3292 | /* |
3293 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
3294 | * and pageblock_default_order() are unused as pageblock_order is set | |
3295 | * at compile-time. See include/linux/pageblock-flags.h for the values of | |
3296 | * pageblock_order based on the kernel config | |
3297 | */ | |
3298 | static inline int pageblock_default_order(unsigned int order) | |
3299 | { | |
3300 | return MAX_ORDER-1; | |
3301 | } | |
d9c23400 MG |
3302 | #define set_pageblock_order(x) do {} while (0) |
3303 | ||
3304 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
3305 | ||
1da177e4 LT |
3306 | /* |
3307 | * Set up the zone data structures: | |
3308 | * - mark all pages reserved | |
3309 | * - mark all memory queues empty | |
3310 | * - clear the memory bitmaps | |
3311 | */ | |
86356ab1 | 3312 | static void __meminit free_area_init_core(struct pglist_data *pgdat, |
1da177e4 LT |
3313 | unsigned long *zones_size, unsigned long *zholes_size) |
3314 | { | |
2f1b6248 | 3315 | enum zone_type j; |
ed8ece2e | 3316 | int nid = pgdat->node_id; |
1da177e4 | 3317 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
718127cc | 3318 | int ret; |
1da177e4 | 3319 | |
208d54e5 | 3320 | pgdat_resize_init(pgdat); |
1da177e4 LT |
3321 | pgdat->nr_zones = 0; |
3322 | init_waitqueue_head(&pgdat->kswapd_wait); | |
3323 | pgdat->kswapd_max_order = 0; | |
3324 | ||
3325 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
3326 | struct zone *zone = pgdat->node_zones + j; | |
0e0b864e | 3327 | unsigned long size, realsize, memmap_pages; |
1da177e4 | 3328 | |
c713216d MG |
3329 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
3330 | realsize = size - zone_absent_pages_in_node(nid, j, | |
3331 | zholes_size); | |
1da177e4 | 3332 | |
0e0b864e MG |
3333 | /* |
3334 | * Adjust realsize so that it accounts for how much memory | |
3335 | * is used by this zone for memmap. This affects the watermark | |
3336 | * and per-cpu initialisations | |
3337 | */ | |
3338 | memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT; | |
3339 | if (realsize >= memmap_pages) { | |
3340 | realsize -= memmap_pages; | |
3341 | printk(KERN_DEBUG | |
3342 | " %s zone: %lu pages used for memmap\n", | |
3343 | zone_names[j], memmap_pages); | |
3344 | } else | |
3345 | printk(KERN_WARNING | |
3346 | " %s zone: %lu pages exceeds realsize %lu\n", | |
3347 | zone_names[j], memmap_pages, realsize); | |
3348 | ||
6267276f CL |
3349 | /* Account for reserved pages */ |
3350 | if (j == 0 && realsize > dma_reserve) { | |
0e0b864e | 3351 | realsize -= dma_reserve; |
6267276f CL |
3352 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
3353 | zone_names[0], dma_reserve); | |
0e0b864e MG |
3354 | } |
3355 | ||
98d2b0eb | 3356 | if (!is_highmem_idx(j)) |
1da177e4 LT |
3357 | nr_kernel_pages += realsize; |
3358 | nr_all_pages += realsize; | |
3359 | ||
3360 | zone->spanned_pages = size; | |
3361 | zone->present_pages = realsize; | |
9614634f | 3362 | #ifdef CONFIG_NUMA |
d5f541ed | 3363 | zone->node = nid; |
8417bba4 | 3364 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
9614634f | 3365 | / 100; |
0ff38490 | 3366 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
9614634f | 3367 | #endif |
1da177e4 LT |
3368 | zone->name = zone_names[j]; |
3369 | spin_lock_init(&zone->lock); | |
3370 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 3371 | zone_seqlock_init(zone); |
1da177e4 | 3372 | zone->zone_pgdat = pgdat; |
1da177e4 | 3373 | |
3bb1a852 | 3374 | zone->prev_priority = DEF_PRIORITY; |
1da177e4 | 3375 | |
ed8ece2e | 3376 | zone_pcp_init(zone); |
1da177e4 LT |
3377 | INIT_LIST_HEAD(&zone->active_list); |
3378 | INIT_LIST_HEAD(&zone->inactive_list); | |
3379 | zone->nr_scan_active = 0; | |
3380 | zone->nr_scan_inactive = 0; | |
2244b95a | 3381 | zap_zone_vm_stats(zone); |
e815af95 | 3382 | zone->flags = 0; |
1da177e4 LT |
3383 | if (!size) |
3384 | continue; | |
3385 | ||
ba72cb8c | 3386 | set_pageblock_order(pageblock_default_order()); |
835c134e | 3387 | setup_usemap(pgdat, zone, size); |
a2f3aa02 DH |
3388 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
3389 | size, MEMMAP_EARLY); | |
718127cc | 3390 | BUG_ON(ret); |
1da177e4 | 3391 | zone_start_pfn += size; |
1da177e4 LT |
3392 | } |
3393 | } | |
3394 | ||
577a32f6 | 3395 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
1da177e4 | 3396 | { |
1da177e4 LT |
3397 | /* Skip empty nodes */ |
3398 | if (!pgdat->node_spanned_pages) | |
3399 | return; | |
3400 | ||
d41dee36 | 3401 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
3402 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
3403 | if (!pgdat->node_mem_map) { | |
e984bb43 | 3404 | unsigned long size, start, end; |
d41dee36 AW |
3405 | struct page *map; |
3406 | ||
e984bb43 BP |
3407 | /* |
3408 | * The zone's endpoints aren't required to be MAX_ORDER | |
3409 | * aligned but the node_mem_map endpoints must be in order | |
3410 | * for the buddy allocator to function correctly. | |
3411 | */ | |
3412 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
3413 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
3414 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
3415 | size = (end - start) * sizeof(struct page); | |
6f167ec7 DH |
3416 | map = alloc_remap(pgdat->node_id, size); |
3417 | if (!map) | |
3418 | map = alloc_bootmem_node(pgdat, size); | |
e984bb43 | 3419 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
1da177e4 | 3420 | } |
12d810c1 | 3421 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3422 | /* |
3423 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
3424 | */ | |
c713216d | 3425 | if (pgdat == NODE_DATA(0)) { |
1da177e4 | 3426 | mem_map = NODE_DATA(0)->node_mem_map; |
c713216d MG |
3427 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
3428 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
467bc461 | 3429 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
c713216d MG |
3430 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
3431 | } | |
1da177e4 | 3432 | #endif |
d41dee36 | 3433 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
3434 | } |
3435 | ||
86356ab1 | 3436 | void __meminit free_area_init_node(int nid, struct pglist_data *pgdat, |
1da177e4 LT |
3437 | unsigned long *zones_size, unsigned long node_start_pfn, |
3438 | unsigned long *zholes_size) | |
3439 | { | |
3440 | pgdat->node_id = nid; | |
3441 | pgdat->node_start_pfn = node_start_pfn; | |
c713216d | 3442 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
1da177e4 LT |
3443 | |
3444 | alloc_node_mem_map(pgdat); | |
3445 | ||
3446 | free_area_init_core(pgdat, zones_size, zholes_size); | |
3447 | } | |
3448 | ||
c713216d | 3449 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
418508c1 MS |
3450 | |
3451 | #if MAX_NUMNODES > 1 | |
3452 | /* | |
3453 | * Figure out the number of possible node ids. | |
3454 | */ | |
3455 | static void __init setup_nr_node_ids(void) | |
3456 | { | |
3457 | unsigned int node; | |
3458 | unsigned int highest = 0; | |
3459 | ||
3460 | for_each_node_mask(node, node_possible_map) | |
3461 | highest = node; | |
3462 | nr_node_ids = highest + 1; | |
3463 | } | |
3464 | #else | |
3465 | static inline void setup_nr_node_ids(void) | |
3466 | { | |
3467 | } | |
3468 | #endif | |
3469 | ||
c713216d MG |
3470 | /** |
3471 | * add_active_range - Register a range of PFNs backed by physical memory | |
3472 | * @nid: The node ID the range resides on | |
3473 | * @start_pfn: The start PFN of the available physical memory | |
3474 | * @end_pfn: The end PFN of the available physical memory | |
3475 | * | |
3476 | * These ranges are stored in an early_node_map[] and later used by | |
3477 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
3478 | * range spans a memory hole, it is up to the architecture to ensure | |
3479 | * the memory is not freed by the bootmem allocator. If possible | |
3480 | * the range being registered will be merged with existing ranges. | |
3481 | */ | |
3482 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
3483 | unsigned long end_pfn) | |
3484 | { | |
3485 | int i; | |
3486 | ||
3487 | printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) " | |
3488 | "%d entries of %d used\n", | |
3489 | nid, start_pfn, end_pfn, | |
3490 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
3491 | ||
3492 | /* Merge with existing active regions if possible */ | |
3493 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3494 | if (early_node_map[i].nid != nid) | |
3495 | continue; | |
3496 | ||
3497 | /* Skip if an existing region covers this new one */ | |
3498 | if (start_pfn >= early_node_map[i].start_pfn && | |
3499 | end_pfn <= early_node_map[i].end_pfn) | |
3500 | return; | |
3501 | ||
3502 | /* Merge forward if suitable */ | |
3503 | if (start_pfn <= early_node_map[i].end_pfn && | |
3504 | end_pfn > early_node_map[i].end_pfn) { | |
3505 | early_node_map[i].end_pfn = end_pfn; | |
3506 | return; | |
3507 | } | |
3508 | ||
3509 | /* Merge backward if suitable */ | |
3510 | if (start_pfn < early_node_map[i].end_pfn && | |
3511 | end_pfn >= early_node_map[i].start_pfn) { | |
3512 | early_node_map[i].start_pfn = start_pfn; | |
3513 | return; | |
3514 | } | |
3515 | } | |
3516 | ||
3517 | /* Check that early_node_map is large enough */ | |
3518 | if (i >= MAX_ACTIVE_REGIONS) { | |
3519 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
3520 | MAX_ACTIVE_REGIONS); | |
3521 | return; | |
3522 | } | |
3523 | ||
3524 | early_node_map[i].nid = nid; | |
3525 | early_node_map[i].start_pfn = start_pfn; | |
3526 | early_node_map[i].end_pfn = end_pfn; | |
3527 | nr_nodemap_entries = i + 1; | |
3528 | } | |
3529 | ||
3530 | /** | |
3531 | * shrink_active_range - Shrink an existing registered range of PFNs | |
3532 | * @nid: The node id the range is on that should be shrunk | |
3533 | * @old_end_pfn: The old end PFN of the range | |
3534 | * @new_end_pfn: The new PFN of the range | |
3535 | * | |
3536 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
3537 | * The map is kept at the end physical page range that has already been | |
3538 | * registered with add_active_range(). This function allows an arch to shrink | |
3539 | * an existing registered range. | |
3540 | */ | |
3541 | void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn, | |
3542 | unsigned long new_end_pfn) | |
3543 | { | |
3544 | int i; | |
3545 | ||
3546 | /* Find the old active region end and shrink */ | |
3547 | for_each_active_range_index_in_nid(i, nid) | |
3548 | if (early_node_map[i].end_pfn == old_end_pfn) { | |
3549 | early_node_map[i].end_pfn = new_end_pfn; | |
3550 | break; | |
3551 | } | |
3552 | } | |
3553 | ||
3554 | /** | |
3555 | * remove_all_active_ranges - Remove all currently registered regions | |
88ca3b94 | 3556 | * |
c713216d MG |
3557 | * During discovery, it may be found that a table like SRAT is invalid |
3558 | * and an alternative discovery method must be used. This function removes | |
3559 | * all currently registered regions. | |
3560 | */ | |
88ca3b94 | 3561 | void __init remove_all_active_ranges(void) |
c713216d MG |
3562 | { |
3563 | memset(early_node_map, 0, sizeof(early_node_map)); | |
3564 | nr_nodemap_entries = 0; | |
fb01439c MG |
3565 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE |
3566 | memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn)); | |
3567 | memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn)); | |
3568 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ | |
c713216d MG |
3569 | } |
3570 | ||
3571 | /* Compare two active node_active_regions */ | |
3572 | static int __init cmp_node_active_region(const void *a, const void *b) | |
3573 | { | |
3574 | struct node_active_region *arange = (struct node_active_region *)a; | |
3575 | struct node_active_region *brange = (struct node_active_region *)b; | |
3576 | ||
3577 | /* Done this way to avoid overflows */ | |
3578 | if (arange->start_pfn > brange->start_pfn) | |
3579 | return 1; | |
3580 | if (arange->start_pfn < brange->start_pfn) | |
3581 | return -1; | |
3582 | ||
3583 | return 0; | |
3584 | } | |
3585 | ||
3586 | /* sort the node_map by start_pfn */ | |
3587 | static void __init sort_node_map(void) | |
3588 | { | |
3589 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
3590 | sizeof(struct node_active_region), | |
3591 | cmp_node_active_region, NULL); | |
3592 | } | |
3593 | ||
a6af2bc3 | 3594 | /* Find the lowest pfn for a node */ |
c713216d MG |
3595 | unsigned long __init find_min_pfn_for_node(unsigned long nid) |
3596 | { | |
3597 | int i; | |
a6af2bc3 | 3598 | unsigned long min_pfn = ULONG_MAX; |
1abbfb41 | 3599 | |
c713216d MG |
3600 | /* Assuming a sorted map, the first range found has the starting pfn */ |
3601 | for_each_active_range_index_in_nid(i, nid) | |
a6af2bc3 | 3602 | min_pfn = min(min_pfn, early_node_map[i].start_pfn); |
c713216d | 3603 | |
a6af2bc3 MG |
3604 | if (min_pfn == ULONG_MAX) { |
3605 | printk(KERN_WARNING | |
3606 | "Could not find start_pfn for node %lu\n", nid); | |
3607 | return 0; | |
3608 | } | |
3609 | ||
3610 | return min_pfn; | |
c713216d MG |
3611 | } |
3612 | ||
3613 | /** | |
3614 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
3615 | * | |
3616 | * It returns the minimum PFN based on information provided via | |
88ca3b94 | 3617 | * add_active_range(). |
c713216d MG |
3618 | */ |
3619 | unsigned long __init find_min_pfn_with_active_regions(void) | |
3620 | { | |
3621 | return find_min_pfn_for_node(MAX_NUMNODES); | |
3622 | } | |
3623 | ||
3624 | /** | |
3625 | * find_max_pfn_with_active_regions - Find the maximum PFN registered | |
3626 | * | |
3627 | * It returns the maximum PFN based on information provided via | |
88ca3b94 | 3628 | * add_active_range(). |
c713216d MG |
3629 | */ |
3630 | unsigned long __init find_max_pfn_with_active_regions(void) | |
3631 | { | |
3632 | int i; | |
3633 | unsigned long max_pfn = 0; | |
3634 | ||
3635 | for (i = 0; i < nr_nodemap_entries; i++) | |
3636 | max_pfn = max(max_pfn, early_node_map[i].end_pfn); | |
3637 | ||
3638 | return max_pfn; | |
3639 | } | |
3640 | ||
37b07e41 LS |
3641 | /* |
3642 | * early_calculate_totalpages() | |
3643 | * Sum pages in active regions for movable zone. | |
3644 | * Populate N_HIGH_MEMORY for calculating usable_nodes. | |
3645 | */ | |
484f51f8 | 3646 | static unsigned long __init early_calculate_totalpages(void) |
7e63efef MG |
3647 | { |
3648 | int i; | |
3649 | unsigned long totalpages = 0; | |
3650 | ||
37b07e41 LS |
3651 | for (i = 0; i < nr_nodemap_entries; i++) { |
3652 | unsigned long pages = early_node_map[i].end_pfn - | |
7e63efef | 3653 | early_node_map[i].start_pfn; |
37b07e41 LS |
3654 | totalpages += pages; |
3655 | if (pages) | |
3656 | node_set_state(early_node_map[i].nid, N_HIGH_MEMORY); | |
3657 | } | |
3658 | return totalpages; | |
7e63efef MG |
3659 | } |
3660 | ||
2a1e274a MG |
3661 | /* |
3662 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
3663 | * is spread evenly between nodes as long as the nodes have enough | |
3664 | * memory. When they don't, some nodes will have more kernelcore than | |
3665 | * others | |
3666 | */ | |
3667 | void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) | |
3668 | { | |
3669 | int i, nid; | |
3670 | unsigned long usable_startpfn; | |
3671 | unsigned long kernelcore_node, kernelcore_remaining; | |
37b07e41 LS |
3672 | unsigned long totalpages = early_calculate_totalpages(); |
3673 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); | |
2a1e274a | 3674 | |
7e63efef MG |
3675 | /* |
3676 | * If movablecore was specified, calculate what size of | |
3677 | * kernelcore that corresponds so that memory usable for | |
3678 | * any allocation type is evenly spread. If both kernelcore | |
3679 | * and movablecore are specified, then the value of kernelcore | |
3680 | * will be used for required_kernelcore if it's greater than | |
3681 | * what movablecore would have allowed. | |
3682 | */ | |
3683 | if (required_movablecore) { | |
7e63efef MG |
3684 | unsigned long corepages; |
3685 | ||
3686 | /* | |
3687 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
3688 | * was requested by the user | |
3689 | */ | |
3690 | required_movablecore = | |
3691 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
3692 | corepages = totalpages - required_movablecore; | |
3693 | ||
3694 | required_kernelcore = max(required_kernelcore, corepages); | |
3695 | } | |
3696 | ||
2a1e274a MG |
3697 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
3698 | if (!required_kernelcore) | |
3699 | return; | |
3700 | ||
3701 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
3702 | find_usable_zone_for_movable(); | |
3703 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; | |
3704 | ||
3705 | restart: | |
3706 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
3707 | kernelcore_node = required_kernelcore / usable_nodes; | |
37b07e41 | 3708 | for_each_node_state(nid, N_HIGH_MEMORY) { |
2a1e274a MG |
3709 | /* |
3710 | * Recalculate kernelcore_node if the division per node | |
3711 | * now exceeds what is necessary to satisfy the requested | |
3712 | * amount of memory for the kernel | |
3713 | */ | |
3714 | if (required_kernelcore < kernelcore_node) | |
3715 | kernelcore_node = required_kernelcore / usable_nodes; | |
3716 | ||
3717 | /* | |
3718 | * As the map is walked, we track how much memory is usable | |
3719 | * by the kernel using kernelcore_remaining. When it is | |
3720 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
3721 | */ | |
3722 | kernelcore_remaining = kernelcore_node; | |
3723 | ||
3724 | /* Go through each range of PFNs within this node */ | |
3725 | for_each_active_range_index_in_nid(i, nid) { | |
3726 | unsigned long start_pfn, end_pfn; | |
3727 | unsigned long size_pages; | |
3728 | ||
3729 | start_pfn = max(early_node_map[i].start_pfn, | |
3730 | zone_movable_pfn[nid]); | |
3731 | end_pfn = early_node_map[i].end_pfn; | |
3732 | if (start_pfn >= end_pfn) | |
3733 | continue; | |
3734 | ||
3735 | /* Account for what is only usable for kernelcore */ | |
3736 | if (start_pfn < usable_startpfn) { | |
3737 | unsigned long kernel_pages; | |
3738 | kernel_pages = min(end_pfn, usable_startpfn) | |
3739 | - start_pfn; | |
3740 | ||
3741 | kernelcore_remaining -= min(kernel_pages, | |
3742 | kernelcore_remaining); | |
3743 | required_kernelcore -= min(kernel_pages, | |
3744 | required_kernelcore); | |
3745 | ||
3746 | /* Continue if range is now fully accounted */ | |
3747 | if (end_pfn <= usable_startpfn) { | |
3748 | ||
3749 | /* | |
3750 | * Push zone_movable_pfn to the end so | |
3751 | * that if we have to rebalance | |
3752 | * kernelcore across nodes, we will | |
3753 | * not double account here | |
3754 | */ | |
3755 | zone_movable_pfn[nid] = end_pfn; | |
3756 | continue; | |
3757 | } | |
3758 | start_pfn = usable_startpfn; | |
3759 | } | |
3760 | ||
3761 | /* | |
3762 | * The usable PFN range for ZONE_MOVABLE is from | |
3763 | * start_pfn->end_pfn. Calculate size_pages as the | |
3764 | * number of pages used as kernelcore | |
3765 | */ | |
3766 | size_pages = end_pfn - start_pfn; | |
3767 | if (size_pages > kernelcore_remaining) | |
3768 | size_pages = kernelcore_remaining; | |
3769 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
3770 | ||
3771 | /* | |
3772 | * Some kernelcore has been met, update counts and | |
3773 | * break if the kernelcore for this node has been | |
3774 | * satisified | |
3775 | */ | |
3776 | required_kernelcore -= min(required_kernelcore, | |
3777 | size_pages); | |
3778 | kernelcore_remaining -= size_pages; | |
3779 | if (!kernelcore_remaining) | |
3780 | break; | |
3781 | } | |
3782 | } | |
3783 | ||
3784 | /* | |
3785 | * If there is still required_kernelcore, we do another pass with one | |
3786 | * less node in the count. This will push zone_movable_pfn[nid] further | |
3787 | * along on the nodes that still have memory until kernelcore is | |
3788 | * satisified | |
3789 | */ | |
3790 | usable_nodes--; | |
3791 | if (usable_nodes && required_kernelcore > usable_nodes) | |
3792 | goto restart; | |
3793 | ||
3794 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
3795 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
3796 | zone_movable_pfn[nid] = | |
3797 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
3798 | } | |
3799 | ||
37b07e41 LS |
3800 | /* Any regular memory on that node ? */ |
3801 | static void check_for_regular_memory(pg_data_t *pgdat) | |
3802 | { | |
3803 | #ifdef CONFIG_HIGHMEM | |
3804 | enum zone_type zone_type; | |
3805 | ||
3806 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { | |
3807 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
3808 | if (zone->present_pages) | |
3809 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
3810 | } | |
3811 | #endif | |
3812 | } | |
3813 | ||
c713216d MG |
3814 | /** |
3815 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
88ca3b94 | 3816 | * @max_zone_pfn: an array of max PFNs for each zone |
c713216d MG |
3817 | * |
3818 | * This will call free_area_init_node() for each active node in the system. | |
3819 | * Using the page ranges provided by add_active_range(), the size of each | |
3820 | * zone in each node and their holes is calculated. If the maximum PFN | |
3821 | * between two adjacent zones match, it is assumed that the zone is empty. | |
3822 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
3823 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
3824 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
3825 | * at arch_max_dma_pfn. | |
3826 | */ | |
3827 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
3828 | { | |
3829 | unsigned long nid; | |
3830 | enum zone_type i; | |
3831 | ||
a6af2bc3 MG |
3832 | /* Sort early_node_map as initialisation assumes it is sorted */ |
3833 | sort_node_map(); | |
3834 | ||
c713216d MG |
3835 | /* Record where the zone boundaries are */ |
3836 | memset(arch_zone_lowest_possible_pfn, 0, | |
3837 | sizeof(arch_zone_lowest_possible_pfn)); | |
3838 | memset(arch_zone_highest_possible_pfn, 0, | |
3839 | sizeof(arch_zone_highest_possible_pfn)); | |
3840 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
3841 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
3842 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2a1e274a MG |
3843 | if (i == ZONE_MOVABLE) |
3844 | continue; | |
c713216d MG |
3845 | arch_zone_lowest_possible_pfn[i] = |
3846 | arch_zone_highest_possible_pfn[i-1]; | |
3847 | arch_zone_highest_possible_pfn[i] = | |
3848 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
3849 | } | |
2a1e274a MG |
3850 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
3851 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
3852 | ||
3853 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
3854 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
3855 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); | |
c713216d | 3856 | |
c713216d MG |
3857 | /* Print out the zone ranges */ |
3858 | printk("Zone PFN ranges:\n"); | |
2a1e274a MG |
3859 | for (i = 0; i < MAX_NR_ZONES; i++) { |
3860 | if (i == ZONE_MOVABLE) | |
3861 | continue; | |
c713216d MG |
3862 | printk(" %-8s %8lu -> %8lu\n", |
3863 | zone_names[i], | |
3864 | arch_zone_lowest_possible_pfn[i], | |
3865 | arch_zone_highest_possible_pfn[i]); | |
2a1e274a MG |
3866 | } |
3867 | ||
3868 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
3869 | printk("Movable zone start PFN for each node\n"); | |
3870 | for (i = 0; i < MAX_NUMNODES; i++) { | |
3871 | if (zone_movable_pfn[i]) | |
3872 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); | |
3873 | } | |
c713216d MG |
3874 | |
3875 | /* Print out the early_node_map[] */ | |
3876 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
3877 | for (i = 0; i < nr_nodemap_entries; i++) | |
3878 | printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid, | |
3879 | early_node_map[i].start_pfn, | |
3880 | early_node_map[i].end_pfn); | |
3881 | ||
3882 | /* Initialise every node */ | |
8ef82866 | 3883 | setup_nr_node_ids(); |
c713216d MG |
3884 | for_each_online_node(nid) { |
3885 | pg_data_t *pgdat = NODE_DATA(nid); | |
3886 | free_area_init_node(nid, pgdat, NULL, | |
3887 | find_min_pfn_for_node(nid), NULL); | |
37b07e41 LS |
3888 | |
3889 | /* Any memory on that node */ | |
3890 | if (pgdat->node_present_pages) | |
3891 | node_set_state(nid, N_HIGH_MEMORY); | |
3892 | check_for_regular_memory(pgdat); | |
c713216d MG |
3893 | } |
3894 | } | |
2a1e274a | 3895 | |
7e63efef | 3896 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
2a1e274a MG |
3897 | { |
3898 | unsigned long long coremem; | |
3899 | if (!p) | |
3900 | return -EINVAL; | |
3901 | ||
3902 | coremem = memparse(p, &p); | |
7e63efef | 3903 | *core = coremem >> PAGE_SHIFT; |
2a1e274a | 3904 | |
7e63efef | 3905 | /* Paranoid check that UL is enough for the coremem value */ |
2a1e274a MG |
3906 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
3907 | ||
3908 | return 0; | |
3909 | } | |
ed7ed365 | 3910 | |
7e63efef MG |
3911 | /* |
3912 | * kernelcore=size sets the amount of memory for use for allocations that | |
3913 | * cannot be reclaimed or migrated. | |
3914 | */ | |
3915 | static int __init cmdline_parse_kernelcore(char *p) | |
3916 | { | |
3917 | return cmdline_parse_core(p, &required_kernelcore); | |
3918 | } | |
3919 | ||
3920 | /* | |
3921 | * movablecore=size sets the amount of memory for use for allocations that | |
3922 | * can be reclaimed or migrated. | |
3923 | */ | |
3924 | static int __init cmdline_parse_movablecore(char *p) | |
3925 | { | |
3926 | return cmdline_parse_core(p, &required_movablecore); | |
3927 | } | |
3928 | ||
ed7ed365 | 3929 | early_param("kernelcore", cmdline_parse_kernelcore); |
7e63efef | 3930 | early_param("movablecore", cmdline_parse_movablecore); |
ed7ed365 | 3931 | |
c713216d MG |
3932 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
3933 | ||
0e0b864e | 3934 | /** |
88ca3b94 RD |
3935 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
3936 | * @new_dma_reserve: The number of pages to mark reserved | |
0e0b864e MG |
3937 | * |
3938 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
3939 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
3940 | * and other unfreeable allocations which can skew the watermarks badly. This | |
88ca3b94 RD |
3941 | * function may optionally be used to account for unfreeable pages in the |
3942 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
3943 | * smaller per-cpu batchsize. | |
0e0b864e MG |
3944 | */ |
3945 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
3946 | { | |
3947 | dma_reserve = new_dma_reserve; | |
3948 | } | |
3949 | ||
93b7504e | 3950 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
3951 | static bootmem_data_t contig_bootmem_data; |
3952 | struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data }; | |
3953 | ||
3954 | EXPORT_SYMBOL(contig_page_data); | |
93b7504e | 3955 | #endif |
1da177e4 LT |
3956 | |
3957 | void __init free_area_init(unsigned long *zones_size) | |
3958 | { | |
93b7504e | 3959 | free_area_init_node(0, NODE_DATA(0), zones_size, |
1da177e4 LT |
3960 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
3961 | } | |
1da177e4 | 3962 | |
1da177e4 LT |
3963 | static int page_alloc_cpu_notify(struct notifier_block *self, |
3964 | unsigned long action, void *hcpu) | |
3965 | { | |
3966 | int cpu = (unsigned long)hcpu; | |
1da177e4 | 3967 | |
8bb78442 | 3968 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
9f8f2172 CL |
3969 | drain_pages(cpu); |
3970 | ||
3971 | /* | |
3972 | * Spill the event counters of the dead processor | |
3973 | * into the current processors event counters. | |
3974 | * This artificially elevates the count of the current | |
3975 | * processor. | |
3976 | */ | |
f8891e5e | 3977 | vm_events_fold_cpu(cpu); |
9f8f2172 CL |
3978 | |
3979 | /* | |
3980 | * Zero the differential counters of the dead processor | |
3981 | * so that the vm statistics are consistent. | |
3982 | * | |
3983 | * This is only okay since the processor is dead and cannot | |
3984 | * race with what we are doing. | |
3985 | */ | |
2244b95a | 3986 | refresh_cpu_vm_stats(cpu); |
1da177e4 LT |
3987 | } |
3988 | return NOTIFY_OK; | |
3989 | } | |
1da177e4 LT |
3990 | |
3991 | void __init page_alloc_init(void) | |
3992 | { | |
3993 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
3994 | } | |
3995 | ||
cb45b0e9 HA |
3996 | /* |
3997 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
3998 | * or min_free_kbytes changes. | |
3999 | */ | |
4000 | static void calculate_totalreserve_pages(void) | |
4001 | { | |
4002 | struct pglist_data *pgdat; | |
4003 | unsigned long reserve_pages = 0; | |
2f6726e5 | 4004 | enum zone_type i, j; |
cb45b0e9 HA |
4005 | |
4006 | for_each_online_pgdat(pgdat) { | |
4007 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
4008 | struct zone *zone = pgdat->node_zones + i; | |
4009 | unsigned long max = 0; | |
4010 | ||
4011 | /* Find valid and maximum lowmem_reserve in the zone */ | |
4012 | for (j = i; j < MAX_NR_ZONES; j++) { | |
4013 | if (zone->lowmem_reserve[j] > max) | |
4014 | max = zone->lowmem_reserve[j]; | |
4015 | } | |
4016 | ||
4017 | /* we treat pages_high as reserved pages. */ | |
4018 | max += zone->pages_high; | |
4019 | ||
4020 | if (max > zone->present_pages) | |
4021 | max = zone->present_pages; | |
4022 | reserve_pages += max; | |
4023 | } | |
4024 | } | |
4025 | totalreserve_pages = reserve_pages; | |
4026 | } | |
4027 | ||
1da177e4 LT |
4028 | /* |
4029 | * setup_per_zone_lowmem_reserve - called whenever | |
4030 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
4031 | * has a correct pages reserved value, so an adequate number of | |
4032 | * pages are left in the zone after a successful __alloc_pages(). | |
4033 | */ | |
4034 | static void setup_per_zone_lowmem_reserve(void) | |
4035 | { | |
4036 | struct pglist_data *pgdat; | |
2f6726e5 | 4037 | enum zone_type j, idx; |
1da177e4 | 4038 | |
ec936fc5 | 4039 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
4040 | for (j = 0; j < MAX_NR_ZONES; j++) { |
4041 | struct zone *zone = pgdat->node_zones + j; | |
4042 | unsigned long present_pages = zone->present_pages; | |
4043 | ||
4044 | zone->lowmem_reserve[j] = 0; | |
4045 | ||
2f6726e5 CL |
4046 | idx = j; |
4047 | while (idx) { | |
1da177e4 LT |
4048 | struct zone *lower_zone; |
4049 | ||
2f6726e5 CL |
4050 | idx--; |
4051 | ||
1da177e4 LT |
4052 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
4053 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
4054 | ||
4055 | lower_zone = pgdat->node_zones + idx; | |
4056 | lower_zone->lowmem_reserve[j] = present_pages / | |
4057 | sysctl_lowmem_reserve_ratio[idx]; | |
4058 | present_pages += lower_zone->present_pages; | |
4059 | } | |
4060 | } | |
4061 | } | |
cb45b0e9 HA |
4062 | |
4063 | /* update totalreserve_pages */ | |
4064 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4065 | } |
4066 | ||
88ca3b94 RD |
4067 | /** |
4068 | * setup_per_zone_pages_min - called when min_free_kbytes changes. | |
4069 | * | |
4070 | * Ensures that the pages_{min,low,high} values for each zone are set correctly | |
4071 | * with respect to min_free_kbytes. | |
1da177e4 | 4072 | */ |
3947be19 | 4073 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
4074 | { |
4075 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
4076 | unsigned long lowmem_pages = 0; | |
4077 | struct zone *zone; | |
4078 | unsigned long flags; | |
4079 | ||
4080 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
4081 | for_each_zone(zone) { | |
4082 | if (!is_highmem(zone)) | |
4083 | lowmem_pages += zone->present_pages; | |
4084 | } | |
4085 | ||
4086 | for_each_zone(zone) { | |
ac924c60 AM |
4087 | u64 tmp; |
4088 | ||
1da177e4 | 4089 | spin_lock_irqsave(&zone->lru_lock, flags); |
ac924c60 AM |
4090 | tmp = (u64)pages_min * zone->present_pages; |
4091 | do_div(tmp, lowmem_pages); | |
1da177e4 LT |
4092 | if (is_highmem(zone)) { |
4093 | /* | |
669ed175 NP |
4094 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
4095 | * need highmem pages, so cap pages_min to a small | |
4096 | * value here. | |
4097 | * | |
4098 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
4099 | * deltas controls asynch page reclaim, and so should | |
4100 | * not be capped for highmem. | |
1da177e4 LT |
4101 | */ |
4102 | int min_pages; | |
4103 | ||
4104 | min_pages = zone->present_pages / 1024; | |
4105 | if (min_pages < SWAP_CLUSTER_MAX) | |
4106 | min_pages = SWAP_CLUSTER_MAX; | |
4107 | if (min_pages > 128) | |
4108 | min_pages = 128; | |
4109 | zone->pages_min = min_pages; | |
4110 | } else { | |
669ed175 NP |
4111 | /* |
4112 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
4113 | * proportionate to the zone's size. |
4114 | */ | |
669ed175 | 4115 | zone->pages_min = tmp; |
1da177e4 LT |
4116 | } |
4117 | ||
ac924c60 AM |
4118 | zone->pages_low = zone->pages_min + (tmp >> 2); |
4119 | zone->pages_high = zone->pages_min + (tmp >> 1); | |
56fd56b8 | 4120 | setup_zone_migrate_reserve(zone); |
1da177e4 LT |
4121 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
4122 | } | |
cb45b0e9 HA |
4123 | |
4124 | /* update totalreserve_pages */ | |
4125 | calculate_totalreserve_pages(); | |
1da177e4 LT |
4126 | } |
4127 | ||
4128 | /* | |
4129 | * Initialise min_free_kbytes. | |
4130 | * | |
4131 | * For small machines we want it small (128k min). For large machines | |
4132 | * we want it large (64MB max). But it is not linear, because network | |
4133 | * bandwidth does not increase linearly with machine size. We use | |
4134 | * | |
4135 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
4136 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
4137 | * | |
4138 | * which yields | |
4139 | * | |
4140 | * 16MB: 512k | |
4141 | * 32MB: 724k | |
4142 | * 64MB: 1024k | |
4143 | * 128MB: 1448k | |
4144 | * 256MB: 2048k | |
4145 | * 512MB: 2896k | |
4146 | * 1024MB: 4096k | |
4147 | * 2048MB: 5792k | |
4148 | * 4096MB: 8192k | |
4149 | * 8192MB: 11584k | |
4150 | * 16384MB: 16384k | |
4151 | */ | |
4152 | static int __init init_per_zone_pages_min(void) | |
4153 | { | |
4154 | unsigned long lowmem_kbytes; | |
4155 | ||
4156 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
4157 | ||
4158 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
4159 | if (min_free_kbytes < 128) | |
4160 | min_free_kbytes = 128; | |
4161 | if (min_free_kbytes > 65536) | |
4162 | min_free_kbytes = 65536; | |
4163 | setup_per_zone_pages_min(); | |
4164 | setup_per_zone_lowmem_reserve(); | |
4165 | return 0; | |
4166 | } | |
4167 | module_init(init_per_zone_pages_min) | |
4168 | ||
4169 | /* | |
4170 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
4171 | * that we can call two helper functions whenever min_free_kbytes | |
4172 | * changes. | |
4173 | */ | |
4174 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
4175 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4176 | { | |
4177 | proc_dointvec(table, write, file, buffer, length, ppos); | |
3b1d92c5 MG |
4178 | if (write) |
4179 | setup_per_zone_pages_min(); | |
1da177e4 LT |
4180 | return 0; |
4181 | } | |
4182 | ||
9614634f CL |
4183 | #ifdef CONFIG_NUMA |
4184 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
4185 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4186 | { | |
4187 | struct zone *zone; | |
4188 | int rc; | |
4189 | ||
4190 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4191 | if (rc) | |
4192 | return rc; | |
4193 | ||
4194 | for_each_zone(zone) | |
8417bba4 | 4195 | zone->min_unmapped_pages = (zone->present_pages * |
9614634f CL |
4196 | sysctl_min_unmapped_ratio) / 100; |
4197 | return 0; | |
4198 | } | |
0ff38490 CL |
4199 | |
4200 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
4201 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4202 | { | |
4203 | struct zone *zone; | |
4204 | int rc; | |
4205 | ||
4206 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4207 | if (rc) | |
4208 | return rc; | |
4209 | ||
4210 | for_each_zone(zone) | |
4211 | zone->min_slab_pages = (zone->present_pages * | |
4212 | sysctl_min_slab_ratio) / 100; | |
4213 | return 0; | |
4214 | } | |
9614634f CL |
4215 | #endif |
4216 | ||
1da177e4 LT |
4217 | /* |
4218 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
4219 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
4220 | * whenever sysctl_lowmem_reserve_ratio changes. | |
4221 | * | |
4222 | * The reserve ratio obviously has absolutely no relation with the | |
4223 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
4224 | * if in function of the boot time zone sizes. | |
4225 | */ | |
4226 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
4227 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4228 | { | |
4229 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4230 | setup_per_zone_lowmem_reserve(); | |
4231 | return 0; | |
4232 | } | |
4233 | ||
8ad4b1fb RS |
4234 | /* |
4235 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
4236 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
4237 | * can have before it gets flushed back to buddy allocator. | |
4238 | */ | |
4239 | ||
4240 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
4241 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
4242 | { | |
4243 | struct zone *zone; | |
4244 | unsigned int cpu; | |
4245 | int ret; | |
4246 | ||
4247 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
4248 | if (!write || (ret == -EINVAL)) | |
4249 | return ret; | |
4250 | for_each_zone(zone) { | |
4251 | for_each_online_cpu(cpu) { | |
4252 | unsigned long high; | |
4253 | high = zone->present_pages / percpu_pagelist_fraction; | |
4254 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
4255 | } | |
4256 | } | |
4257 | return 0; | |
4258 | } | |
4259 | ||
f034b5d4 | 4260 | int hashdist = HASHDIST_DEFAULT; |
1da177e4 LT |
4261 | |
4262 | #ifdef CONFIG_NUMA | |
4263 | static int __init set_hashdist(char *str) | |
4264 | { | |
4265 | if (!str) | |
4266 | return 0; | |
4267 | hashdist = simple_strtoul(str, &str, 0); | |
4268 | return 1; | |
4269 | } | |
4270 | __setup("hashdist=", set_hashdist); | |
4271 | #endif | |
4272 | ||
4273 | /* | |
4274 | * allocate a large system hash table from bootmem | |
4275 | * - it is assumed that the hash table must contain an exact power-of-2 | |
4276 | * quantity of entries | |
4277 | * - limit is the number of hash buckets, not the total allocation size | |
4278 | */ | |
4279 | void *__init alloc_large_system_hash(const char *tablename, | |
4280 | unsigned long bucketsize, | |
4281 | unsigned long numentries, | |
4282 | int scale, | |
4283 | int flags, | |
4284 | unsigned int *_hash_shift, | |
4285 | unsigned int *_hash_mask, | |
4286 | unsigned long limit) | |
4287 | { | |
4288 | unsigned long long max = limit; | |
4289 | unsigned long log2qty, size; | |
4290 | void *table = NULL; | |
4291 | ||
4292 | /* allow the kernel cmdline to have a say */ | |
4293 | if (!numentries) { | |
4294 | /* round applicable memory size up to nearest megabyte */ | |
04903664 | 4295 | numentries = nr_kernel_pages; |
1da177e4 LT |
4296 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
4297 | numentries >>= 20 - PAGE_SHIFT; | |
4298 | numentries <<= 20 - PAGE_SHIFT; | |
4299 | ||
4300 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
4301 | if (scale > PAGE_SHIFT) | |
4302 | numentries >>= (scale - PAGE_SHIFT); | |
4303 | else | |
4304 | numentries <<= (PAGE_SHIFT - scale); | |
9ab37b8f PM |
4305 | |
4306 | /* Make sure we've got at least a 0-order allocation.. */ | |
4307 | if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
4308 | numentries = PAGE_SIZE / bucketsize; | |
1da177e4 | 4309 | } |
6e692ed3 | 4310 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
4311 | |
4312 | /* limit allocation size to 1/16 total memory by default */ | |
4313 | if (max == 0) { | |
4314 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
4315 | do_div(max, bucketsize); | |
4316 | } | |
4317 | ||
4318 | if (numentries > max) | |
4319 | numentries = max; | |
4320 | ||
f0d1b0b3 | 4321 | log2qty = ilog2(numentries); |
1da177e4 LT |
4322 | |
4323 | do { | |
4324 | size = bucketsize << log2qty; | |
4325 | if (flags & HASH_EARLY) | |
4326 | table = alloc_bootmem(size); | |
4327 | else if (hashdist) | |
4328 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
4329 | else { | |
4330 | unsigned long order; | |
4331 | for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++) | |
4332 | ; | |
4333 | table = (void*) __get_free_pages(GFP_ATOMIC, order); | |
1037b83b ED |
4334 | /* |
4335 | * If bucketsize is not a power-of-two, we may free | |
4336 | * some pages at the end of hash table. | |
4337 | */ | |
4338 | if (table) { | |
4339 | unsigned long alloc_end = (unsigned long)table + | |
4340 | (PAGE_SIZE << order); | |
4341 | unsigned long used = (unsigned long)table + | |
4342 | PAGE_ALIGN(size); | |
4343 | split_page(virt_to_page(table), order); | |
4344 | while (used < alloc_end) { | |
4345 | free_page(used); | |
4346 | used += PAGE_SIZE; | |
4347 | } | |
4348 | } | |
1da177e4 LT |
4349 | } |
4350 | } while (!table && size > PAGE_SIZE && --log2qty); | |
4351 | ||
4352 | if (!table) | |
4353 | panic("Failed to allocate %s hash table\n", tablename); | |
4354 | ||
b49ad484 | 4355 | printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n", |
1da177e4 LT |
4356 | tablename, |
4357 | (1U << log2qty), | |
f0d1b0b3 | 4358 | ilog2(size) - PAGE_SHIFT, |
1da177e4 LT |
4359 | size); |
4360 | ||
4361 | if (_hash_shift) | |
4362 | *_hash_shift = log2qty; | |
4363 | if (_hash_mask) | |
4364 | *_hash_mask = (1 << log2qty) - 1; | |
4365 | ||
4366 | return table; | |
4367 | } | |
a117e66e KH |
4368 | |
4369 | #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE | |
a117e66e KH |
4370 | struct page *pfn_to_page(unsigned long pfn) |
4371 | { | |
67de6482 | 4372 | return __pfn_to_page(pfn); |
a117e66e KH |
4373 | } |
4374 | unsigned long page_to_pfn(struct page *page) | |
4375 | { | |
67de6482 | 4376 | return __page_to_pfn(page); |
a117e66e | 4377 | } |
a117e66e KH |
4378 | EXPORT_SYMBOL(pfn_to_page); |
4379 | EXPORT_SYMBOL(page_to_pfn); | |
4380 | #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */ | |
6220ec78 | 4381 | |
835c134e MG |
4382 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
4383 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
4384 | unsigned long pfn) | |
4385 | { | |
4386 | #ifdef CONFIG_SPARSEMEM | |
4387 | return __pfn_to_section(pfn)->pageblock_flags; | |
4388 | #else | |
4389 | return zone->pageblock_flags; | |
4390 | #endif /* CONFIG_SPARSEMEM */ | |
4391 | } | |
4392 | ||
4393 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
4394 | { | |
4395 | #ifdef CONFIG_SPARSEMEM | |
4396 | pfn &= (PAGES_PER_SECTION-1); | |
d9c23400 | 4397 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
4398 | #else |
4399 | pfn = pfn - zone->zone_start_pfn; | |
d9c23400 | 4400 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
835c134e MG |
4401 | #endif /* CONFIG_SPARSEMEM */ |
4402 | } | |
4403 | ||
4404 | /** | |
d9c23400 | 4405 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
835c134e MG |
4406 | * @page: The page within the block of interest |
4407 | * @start_bitidx: The first bit of interest to retrieve | |
4408 | * @end_bitidx: The last bit of interest | |
4409 | * returns pageblock_bits flags | |
4410 | */ | |
4411 | unsigned long get_pageblock_flags_group(struct page *page, | |
4412 | int start_bitidx, int end_bitidx) | |
4413 | { | |
4414 | struct zone *zone; | |
4415 | unsigned long *bitmap; | |
4416 | unsigned long pfn, bitidx; | |
4417 | unsigned long flags = 0; | |
4418 | unsigned long value = 1; | |
4419 | ||
4420 | zone = page_zone(page); | |
4421 | pfn = page_to_pfn(page); | |
4422 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4423 | bitidx = pfn_to_bitidx(zone, pfn); | |
4424 | ||
4425 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4426 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
4427 | flags |= value; | |
6220ec78 | 4428 | |
835c134e MG |
4429 | return flags; |
4430 | } | |
4431 | ||
4432 | /** | |
d9c23400 | 4433 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
835c134e MG |
4434 | * @page: The page within the block of interest |
4435 | * @start_bitidx: The first bit of interest | |
4436 | * @end_bitidx: The last bit of interest | |
4437 | * @flags: The flags to set | |
4438 | */ | |
4439 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
4440 | int start_bitidx, int end_bitidx) | |
4441 | { | |
4442 | struct zone *zone; | |
4443 | unsigned long *bitmap; | |
4444 | unsigned long pfn, bitidx; | |
4445 | unsigned long value = 1; | |
4446 | ||
4447 | zone = page_zone(page); | |
4448 | pfn = page_to_pfn(page); | |
4449 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4450 | bitidx = pfn_to_bitidx(zone, pfn); | |
4451 | ||
4452 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4453 | if (flags & value) | |
4454 | __set_bit(bitidx + start_bitidx, bitmap); | |
4455 | else | |
4456 | __clear_bit(bitidx + start_bitidx, bitmap); | |
4457 | } | |
a5d76b54 KH |
4458 | |
4459 | /* | |
4460 | * This is designed as sub function...plz see page_isolation.c also. | |
4461 | * set/clear page block's type to be ISOLATE. | |
4462 | * page allocater never alloc memory from ISOLATE block. | |
4463 | */ | |
4464 | ||
4465 | int set_migratetype_isolate(struct page *page) | |
4466 | { | |
4467 | struct zone *zone; | |
4468 | unsigned long flags; | |
4469 | int ret = -EBUSY; | |
4470 | ||
4471 | zone = page_zone(page); | |
4472 | spin_lock_irqsave(&zone->lock, flags); | |
4473 | /* | |
4474 | * In future, more migrate types will be able to be isolation target. | |
4475 | */ | |
4476 | if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE) | |
4477 | goto out; | |
4478 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); | |
4479 | move_freepages_block(zone, page, MIGRATE_ISOLATE); | |
4480 | ret = 0; | |
4481 | out: | |
4482 | spin_unlock_irqrestore(&zone->lock, flags); | |
4483 | if (!ret) | |
9f8f2172 | 4484 | drain_all_pages(); |
a5d76b54 KH |
4485 | return ret; |
4486 | } | |
4487 | ||
4488 | void unset_migratetype_isolate(struct page *page) | |
4489 | { | |
4490 | struct zone *zone; | |
4491 | unsigned long flags; | |
4492 | zone = page_zone(page); | |
4493 | spin_lock_irqsave(&zone->lock, flags); | |
4494 | if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE) | |
4495 | goto out; | |
4496 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
4497 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
4498 | out: | |
4499 | spin_unlock_irqrestore(&zone->lock, flags); | |
4500 | } | |
0c0e6195 KH |
4501 | |
4502 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
4503 | /* | |
4504 | * All pages in the range must be isolated before calling this. | |
4505 | */ | |
4506 | void | |
4507 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
4508 | { | |
4509 | struct page *page; | |
4510 | struct zone *zone; | |
4511 | int order, i; | |
4512 | unsigned long pfn; | |
4513 | unsigned long flags; | |
4514 | /* find the first valid pfn */ | |
4515 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
4516 | if (pfn_valid(pfn)) | |
4517 | break; | |
4518 | if (pfn == end_pfn) | |
4519 | return; | |
4520 | zone = page_zone(pfn_to_page(pfn)); | |
4521 | spin_lock_irqsave(&zone->lock, flags); | |
4522 | pfn = start_pfn; | |
4523 | while (pfn < end_pfn) { | |
4524 | if (!pfn_valid(pfn)) { | |
4525 | pfn++; | |
4526 | continue; | |
4527 | } | |
4528 | page = pfn_to_page(pfn); | |
4529 | BUG_ON(page_count(page)); | |
4530 | BUG_ON(!PageBuddy(page)); | |
4531 | order = page_order(page); | |
4532 | #ifdef CONFIG_DEBUG_VM | |
4533 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
4534 | pfn, 1 << order, end_pfn); | |
4535 | #endif | |
4536 | list_del(&page->lru); | |
4537 | rmv_page_order(page); | |
4538 | zone->free_area[order].nr_free--; | |
4539 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
4540 | - (1UL << order)); | |
4541 | for (i = 0; i < (1 << order); i++) | |
4542 | SetPageReserved((page+i)); | |
4543 | pfn += (1 << order); | |
4544 | } | |
4545 | spin_unlock_irqrestore(&zone->lock, flags); | |
4546 | } | |
4547 | #endif |